---
source: 'https://howaiworks.ai/blog'
section: blog
count: 311
---

# Blog

> All 311 pages in the Blog section. Each link is the Markdown twin; drop the `.md` for the HTML page.

- [Adobe Acrobat AI: Presentations, Podcasts & Chat](https://howaiworks.ai/blog/adobe-acrobat-ai-presentations-podcasts.md) — updated 2026-01-26
  Adobe announced a major update to Acrobat: generate presentations and podcasts from PDFs, edit documents via AI chat, and collaborate in PDF Spaces.
- [Seven Chokepoints in the AI Chip Supply Chain](https://howaiworks.ai/blog/ai-chip-supply-chain-chokepoints.md) — updated 2026-07-13
  A walk down the chain behind one AI accelerator — design tools, foundry, scanner, memory, packaging, software, power — and why each link has no second source.
- [AirPods Pro 3: AI Live Translation Revolution](https://howaiworks.ai/blog/airpods-pro-3-live-translation.md) — updated 2025-09-09
  AirPods Pro 3 feature AI-powered Live Translation in 9 languages. Discover how Apple Intelligence enables real-time speech translation through earbuds.
- [Alibaba GUI-Owl-1.5 & Mobile-Agent-v3.5: The Next Era of GUI Agents](https://howaiworks.ai/blog/alibaba-gui-owl-1-5-mobile-agent-v3-5.md) — updated 2026-03-09
  Alibaba releases GUI-Owl 1.5, a multi-modal mobile agent that can autonomously navigate and interact with complex smartphone applications through vision.
- [Alibaba's Qoder Coding Agent Gets a Smart Glasses Client](https://howaiworks.ai/blog/alibaba-qoder-glasses-edition.md) — updated 2026-09-04
  Alibaba announced a glasses edition of its Qoder coding platform on September 3, 2026, running on Qwen and Rokid AI glasses with full-duplex voice.
- [Qwen 3.5: Scaling Intelligence in Compact Models](https://howaiworks.ai/blog/alibaba-qwen-3-5-compact-models-announcement.md) — updated 2026-07-21
  Alibaba's new Qwen 3.5 series packs flagship intelligence into compact sizes (0.8B to 9B), featuring native multimodality and enhanced agentic capabilities.
- [Alibaba Qwen 3.5: 1M Token Context and Efficiency](https://howaiworks.ai/blog/alibaba-qwen-3-5-medium-announcement.md) — updated 2026-02-25
  Alibaba announces Qwen 3.5 Medium, an 8B parameter model featuring enhanced reasoning and a 256k context window for professional coding and research tasks.
- [Alibaba Qwen 3.6-Plus: 1M Context Window and Agentic Coding](https://howaiworks.ai/blog/alibaba-qwen-3-6-plus-announcement.md) — updated 2026-04-03
  Alibaba officially unveils Qwen 3.6-Plus, a flagship model featuring a 1-million-token context window and optimized for repository-level agentic coding.
- [Qwen 3.7-Max: Alibaba's Long-Horizon Agent Engine](https://howaiworks.ai/blog/alibaba-qwen-3-7-max-agent-engine.md) — updated 2026-05-21
  Alibaba launches Qwen 3.7-Max, a flagship AI model demonstrating 35-hour autonomous operation, 10x kernel speedup, and cross-agent generalization.
- [Qwen3.8-Flash-Next Runs Locally in 75 GB of RAM](https://howaiworks.ai/blog/alibaba-qwen-3-8-flash-next-local-gguf.md) — updated 2026-08-27
  Alibaba's 125B multimodal MoE is out as Unsloth GGUFs. The 1-bit build fits in about 75 GB of RAM or unified memory, with no GPU VRAM required.
- [Qwen3.8-Max's Open Weights Are Not the Model You Benchmarked](https://howaiworks.ai/blog/alibaba-qwen-3-8-max-open-weights-license.md) — updated 2026-09-06
  Alibaba opened its 2.4T flagship — but the download is text-only, thinking-required and 262K context, under a licence that is not Apache 2.0.
- [Alibaba Launches Qwen App to Challenge ChatGPT](https://howaiworks.ai/blog/alibaba-qwen-app-launch-2025.md) — updated 2025-11-18
  Alibaba launches Qwen app powered by Qwen3, offering free access and competing directly with ChatGPT in the consumer AI market.
- [Qwen-Scope: Alibaba's Open 'X-Ray' for Model Interpretability](https://howaiworks.ai/blog/alibaba-qwen-scope-interpretability-sae.md) — updated 2026-05-01
  Alibaba releases Qwen-Scope, a massive collection of Sparse Autoencoders (SAEs) that allows researchers to 'look inside' Qwen models and steer their behavior.
- [Alibaba Qwen Team Faces Key Departures and Restructuring](https://howaiworks.ai/blog/alibaba-qwen-team-departures-restructuring.md) — updated 2026-03-09
  Alibaba's Qwen team undergoes restructuring following key departures, aiming to accelerate the development of next-generation large language models.
- [The Newest Wan You Can Download Is Still Wan 2.2](https://howaiworks.ai/blog/alibaba-wan-open-weights-stopped-at-2-2.md) — updated 2026-09-06
  Alibaba shipped Wan 2.6, 2.7 and 3.0 with no weights. Only Wan 2.2 is downloadable — and the pages claiming Wan 2.7 is Apache 2.0 are wrong.
- [Nvidia, Microsoft, xAI Lead $40B Deal](https://howaiworks.ai/blog/aligned-data-centers-40-billion-deal-2025.md) — updated 2025-10-16
  Aligned Data Centers secures a $40 billion investment to build the next generation of AI-optimized infrastructure with liquid cooling and sustainable energy.
- [Amazon Nova 2: Four New Models Plus Forge and Act](https://howaiworks.ai/blog/amazon-nova-2-models-announcement-2025.md) — updated 2025-12-02
  Amazon announces Nova 2 model family with Lite, Pro, Sonic, and Omni, plus Nova Forge for custom models and Nova Act for reliable AI agents.
- [Building Personal Knowledge Bases with LLMs: The Karpathy Method](https://howaiworks.ai/blog/andrej-karpathy-llm-knowledge-bases.md) — updated 2026-04-03
  Explore Andrej Karpathy's workflow for using LLMs to incrementally compile and manage a massive personal knowledge base in Obsidian.
- [Karpathy's nanochat and LLM Council: Build Your Own AI](https://howaiworks.ai/blog/andrej-karpathy-nanochat-llm-council.md) — updated 2026-07-24
  nanochat is Andrej Karpathy's ~$100 from-scratch ChatGPT clone, and LLM Council polls several frontier models. What each is and how they work.
- [Ant Group Unveils Ling-1T: Trillion-Param Model](https://howaiworks.ai/blog/ant-group-ling-1t-announcement.md) — updated 2025-10-13
  Ant Group releases Ling-1T, a trillion-parameter open-source AI model with state-of-the-art coding, reasoning, and multimodal capabilities.
- [Ant Group's Ling-2.6-flash: Lean MoE for AI Agents](https://howaiworks.ai/blog/ant-group-ling-2-6-flash-release.md) — updated 2026-04-22
  Ant Group releases Ling-2.6-flash, an efficient 104B MoE model optimized for 'intelligence per token,' agentic workflows, and fast long-context performance.
- [Ant Group Releases LingBot-Depth: A 2.7 TB RGB-D Dataset for Robotics](https://howaiworks.ai/blog/ant-group-lingbot-depth-dataset-huggingface.md) — updated 2026-04-02
  Ant Group has published LingBot-Depth on Hugging Face, a massive 2.7 TB dataset featuring over 3 million RGB-D examples for advancing spatial perception.
- [Anthropic Agent Skills: Customizable AI Tasks](https://howaiworks.ai/blog/anthropic-agent-skills-announcement.md) — updated 2025-10-17
  Anthropic announces Agent Skills, a system that allows Claude to load specialized instructions and resources to improve performance on specific tasks.
- [How AI Transforms Work: Anthropic Research 2025](https://howaiworks.ai/blog/anthropic-ai-transforming-work-research-2025.md) — updated 2025-12-05
  Anthropic research: AI boosts productivity 50%, enables new work types. Insights from 132 engineers on how Claude transforms software engineering work.
- [Assistant Axis: Controlling LLM Character](https://howaiworks.ai/blog/anthropic-assistant-axis-research-2025.md) — updated 2026-01-23
  Anthropic research reveals how LLMs drift between personas. The Assistant Axis stabilizes model behavior and prevents harmful outputs via activation capping.
- [Anthropic Ships Open-Source Commerce Agent Blueprints](https://howaiworks.ai/blog/anthropic-commerce-agents-blueprint-2026.md) — updated 2026-09-04
  Anthropic released Apache-2.0 blueprints for a shopping agent and a merchant agent on Claude, with Visa, Mastercard and Accenture as distribution partners.
- [Context Engineering: AI Agent Optimization Guide](https://howaiworks.ai/blog/anthropic-context-engineering-for-agents.md) — updated 2025-10-01
  Anthropic reveals advanced strategies for managing context in AI agents, from token optimization to long-horizon task handling and multi-agent architectures.
- [Anthropic: 250 Malicious Docs Can Poison Any LLM](https://howaiworks.ai/blog/anthropic-data-poisoning-research-2025.md) — updated 2026-07-08
  Anthropic publishes research on data poisoning, proposing new defense mechanisms to protect large language models from adversarial training data attacks.
- [Anthropic in Talks with Google for Cloud Deal](https://howaiworks.ai/blog/anthropic-google-cloud-deal-2025.md) — updated 2025-10-22
  AI startup Anthropic discusses a multi-billion dollar cloud computing deal with Google to significantly accelerate AI development and infrastructure scaling.
- [OpenAI Hardware Engineer Moves to Anthropic](https://howaiworks.ai/blog/anthropic-hires-clive-chan.md) — updated 2026-06-09
  Anthropic has hired former OpenAI engineer Clive Chan to develop its own AI chips in order to reduce computing costs.
- [Anthropic Interviewer: Tool for AI Impact](https://howaiworks.ai/blog/anthropic-interviewer-research-tool.md) — updated 2025-12-05
  Anthropic releases a new AI interviewer research tool designed to conduct structured interviews and synthesize insights using advanced language understanding.
- [US Orders Anthropic to Suspend Fable 5 and Mythos 5 Access](https://howaiworks.ai/blog/anthropic-suspends-fable-5-mythos-5-2026.md) — updated 2026-06-12
  Anthropic halts access to its latest models, Fable 5 and Mythos 5, after a US government directive citing national security concerns over jailbreak methods.
- [AI Agent Tools: Anthropic Development Guide](https://howaiworks.ai/blog/anthropic-writing-tools-for-agents.md) — updated 2025-09-14
  Discover Anthropic's proven techniques for building high-quality tools that maximize AI agent performance, from prototyping to evaluation and optimization.
- [Antirez: Control the Ideas, Not the Code](https://howaiworks.ai/blog/antirez-control-the-ideas-not-the-code.md) — updated 2026-07-13
  Redis creator antirez says reading AI-generated code line by line wastes a programmer's day. What he proposes instead, and where the argument holds.
- [Apple M5 Chip: Next-Generation AI Performance](https://howaiworks.ai/blog/apple-m5-chip-announcement-2025.md) — updated 2025-10-15
  Apple announces M5 chip with 4x GPU AI performance, Neural Accelerators, and enhanced unified memory for MacBook Pro, iPad Pro, and Apple Vision Pro.
- [Apple Builds Macs for AI Clusters as OpenAI Reportedly Stocks Up](https://howaiworks.ai/blog/apple-mac-ai-hardware-openai-mac-mini-2026.md) — updated 2026-08-31
  Mac revenue hit $10.35B, up ~29%. The Information reports OpenAI bought tens of thousands of Macs for RL, days after Apple pitched new Macs at local AI.
- [Apple Unveils Siri AI Powered by Google Gemini](https://howaiworks.ai/blog/apple-siri-ai-wwdc-2026.md) — updated 2026-06-09
  At WWDC 2026, Apple introduced a completely redesigned version of its voice assistant, Siri AI, featuring visual intelligence and on-device data processing.
- [Awesome DeepSeek Agent: 22 Tools You Can Point at V4](https://howaiworks.ai/blog/awesome-deepseek-agent-integration-guides.md) — updated 2026-07-24
  DeepSeek's official awesome-list ships setup guides for running V4-Pro and V4-Flash inside Claude Code, Cline, Codex, Qwen Code and 18 other agents.
- [Best AI App Builders in 2026](https://howaiworks.ai/blog/best-ai-app-builders.md) — updated 2026-07-10
  Lovable, Bolt, v0, Base44, and Replit Agent all turn prompts into working apps. They differ on portability, cost, and control. Here's how to choose.
- [Best AI Image Model 2026: Price and Quality Have Decoupled](https://howaiworks.ai/blog/best-ai-image-model.md) — updated 2026-08-02
  GPT Image 2 leads the arena, Google's cheapest model beats its priciest, and Midjourney is absent from the top twelve. What the data says in August 2026.
- [Best AI Model for Financial Analysis in 2026](https://howaiworks.ai/blog/best-ai-model-financial-analysis.md) — updated 2026-07-24
  Which AI models lead on finance benchmarks in 2026, why RAG is worth its steep token cost, and why a confidently wrong number should shape your choice.
- [Best AI Model for Legal Writing in 2026](https://howaiworks.ai/blog/best-ai-model-legal-writing.md) — updated 2026-07-24
  Which AI models are strongest for legal drafting and research in 2026 — and why none of them is safe to use without a citation-verification layer.
- [Best AI Model for Medical & Clinical Documentation](https://howaiworks.ai/blog/best-ai-model-medical-documentation.md) — updated 2026-07-24
  Which AI models lead on medical benchmarks in 2026, why a high MedQA score doesn't mean clinical safety, and the HIPAA and verification guardrails you need.
- [The Best AI Video Generator in 2026 Is Probably Not the One You Think](https://howaiworks.ai/blog/best-ai-video-generator.md) — updated 2026-08-02
  Sora is being switched off, Veo 3.1 ranks 11th, and Chinese models hold most of the top ten. What the video arena actually says in August 2026.
- [Best LLM for RAG & Long-Context Summarization](https://howaiworks.ai/blog/best-llm-rag-long-context-summarization.md) — updated 2026-07-20
  Why the needle-in-a-haystack score misleads, which models hold up on multi-fact retrieval in 2026, and when to use RAG over a giant context window.
- [Best Open-Weight LLM for Agentic Coding in 2026](https://howaiworks.ai/blog/best-open-weight-llm-agentic-coding.md) — updated 2026-07-20
  A decision guide to the top open-weight coding models in 2026 — DeepSeek V4, GLM-5.2, Kimi K3, Qwen, and MiniMax M3 — by price, license, and context.
- [Bezos Project Prometheus: AI Initiative Revealed](https://howaiworks.ai/blog/bezos-project-prometheus-2025.md) — updated 2025-11-18
  Jeff Bezos launches Project Prometheus, a major AI initiative focused on developing advanced artificial intelligence technologies for global challenges.
- [ByteDance's Doubao: China's Leading AI Chatbot](https://howaiworks.ai/blog/bytedance-doubao-china-leading-chatbot.md) — updated 2025-10-17
  ByteDance's Doubao surpasses DeepSeek to become China's most popular AI chatbot, ranking fourth globally with user-friendly design and viral social features.
- [Chad IDE: Y Combinator's Brainrot Coding Tool](https://howaiworks.ai/blog/chad-brainrot-ide-y-combinator-2025.md) — updated 2025-11-14
  Y Combinator-backed Chad IDE lets developers gamble, watch TikToks, and play games while AI coding assistants work, sparking debate about productivity.
- [ChatGPT-5.4 Leaks: 2M Context, Full-Res Vision, and Agentic Power](https://howaiworks.ai/blog/chatgpt-5-4-leaks-announcement.md) — updated 2026-03-03
  Recent leaks regarding ChatGPT 5.4 reveal major upgrades in agentic orchestration, native 3D world understanding, and significant efficiency improvements.
- [The Cheapest LLM APIs for High-Volume Work in 2026](https://howaiworks.ai/blog/cheapest-llm-api-high-volume.md) — updated 2026-07-20
  For jobs that run millions of times, cost per task beats cost per token. A 2026 guide to budget LLM APIs, batch discounts, and prompt caching.
- [China Claims 14nm Chips Rival NVIDIA's 4nm](https://howaiworks.ai/blog/china-14nm-chips-rival-nvidia-4nm-2025.md) — updated 2025-12-02
  China's new 14nm processor with 18nm DRAM achieves 120 TFLOPS, outperforming NVIDIA A100 GPUs and addressing memory bandwidth challenges.
- [China Invests $295B in Huawei-Powered AI Infrastructure](https://howaiworks.ai/blog/china-295b-ai-infrastructure-huawei.md) — updated 2026-06-12
  China announces a $295B plan to build a nationwide AI data center network, aiming to replace Nvidia and AMD with domestic Huawei chips.
- [China's Analog Chip 1,000x Faster Than GPUs](https://howaiworks.ai/blog/china-analog-chip-rram-2025.md) — updated 2025-11-17
  Chinese researchers announce a breakthrough in analog computing using RRAM, enabling ultra-low power AI inference for edge devices and wearable technology.
- [China's Tech Giants Race for Agentic Commerce](https://howaiworks.ai/blog/china-tech-giants-agentic-commerce-race.md) — updated 2026-01-27
  Chinese tech giants like Alibaba, Tencent, and ByteDance are racing to build AI-powered agentic commerce super apps.
- [Cici Is Now Dola: ByteDance's Overseas AI Assistant](https://howaiworks.ai/blog/cici-dola-bytedance-ai-assistant.md) — updated 2026-07-21
  Cici was ByteDance's overseas AI assistant, renamed Dola in November 2025. What changed, whether it's safe, and where it's available.
- [Building AI Agents with Claude Agent SDK](https://howaiworks.ai/blog/claude-agent-sdk-building-agents.md) — updated 2026-07-08
  Anthropic releases the Claude Agent SDK, providing a comprehensive toolkit for developers to build, test, and deploy autonomous AI agents with Claude.
- [Claude Code Authentication Errors: OAuth vs API Key](https://howaiworks.ai/blog/claude-code-authentication-errors-oauth-api-key.md) — updated 2026-07-20
  Fix Claude Code login failures, 403 errors, and surprise API charges caused by ANTHROPIC_API_KEY overriding your subscription — with the exact commands to run.
- [Claude Code Auto-Compact: What It Means & How to Control It](https://howaiworks.ai/blog/claude-code-auto-compact-context-management.md) — updated 2026-07-20
  Understand the 'context left until auto-compact' indicator in Claude Code, fix the auto-compact thrashing error, and take manual control of your context window.
- [Claude Code Auto-dream: New Agentic Memory Management](https://howaiworks.ai/blog/claude-code-auto-dream-memory-feature.md) — updated 2026-03-24
  Claude Code introduces Auto-Dream memory, allowing autonomous agents to refine their understanding of codebases during idle time for better accuracy.
- [Mastering Claude Code: Systemic Patterns for Agentic Engineering](https://howaiworks.ai/blog/claude-code-best-practices-repository.md) — updated 2026-07-08
  Anthropic releases an official repository of best practices for Claude Code, featuring advanced strategies for autonomous engineering and multi-file tasks.
- [Visual Masterclass: Mastering Claude Code and Agents](https://howaiworks.ai/blog/claude-code-masterclass-visual-guide.md) — updated 2026-07-08
  Master Anthropic's Claude Code with a comprehensive visual guide covering everything from basic setup to advanced multi-agent workflows and MCP integrations.
- [Claude Code: The Ultimate Resource Collection for Pro Developers](https://howaiworks.ai/blog/claude-code-resource-collection-2026.md) — updated 2026-04-02
  Master Anthropic's Claude Code with this curated selection of repositories, documentation, tutorials, and books.
- [Claude Code Introduces /ultrareview: New Fleet of Bug-Hunters](https://howaiworks.ai/blog/claude-code-ultrareview-agentic-code-analysis.md) — updated 2026-04-23
  Explore the new /ultrareview feature in Claude Code, a cloud-based multi-agent system designed to identify and verify deep-seated bugs before code merges.
- [Claude Code vs Codex vs Antigravity: Picking a Coding Agent](https://howaiworks.ai/blog/claude-code-vs-codex-cli-vs-antigravity-cli.md) — updated 2026-08-02
  Three coding agents, three bets on openness, model lock-in and pricing. What separates them in August 2026 — and why benchmarks don't decide it.
- [Claude Code vs Cursor vs Windsurf: Which Coding Agent?](https://howaiworks.ai/blog/claude-code-vs-cursor-vs-windsurf-which-coding-agent.md) — updated 2026-07-24
  A workflow-first comparison of the three leading AI coding agents in 2026 — their pricing, their philosophies, and which one fits how you actually work.
- [Claude Cowork: AI Agent for macOS and Windows](https://howaiworks.ai/blog/claude-cowork-research-preview.md) — updated 2026-02-16
  Anthropic's Cowork research preview brings Claude Code's agentic capabilities to everyone, now available on Windows with new global instructions.
- [Claude Cowork: AI Assistant for Files & Folders](https://howaiworks.ai/blog/claude-cowork-research-preview-2025.md) — updated 2026-01-12
  Anthropic launches Cowork, a research preview that gives Claude access to your computer files, enabling autonomous task completion beyond coding.
- [Claude Fable 5.1: Same $10/$50, Cache Reads Cut 75%](https://howaiworks.ai/blog/claude-fable-5-1-announcement-2026.md) — updated 2026-09-06
  Anthropic shipped Claude Fable 5.1 and Mythos 5.1 on September 1, 2026. Sticker prices are unchanged; cache reads fall from $1 to $0.25 per million.
- [Claude Fable 5: The Next Generation of Frontier Intelligence](https://howaiworks.ai/blog/claude-fable-5-announcement-2026.md) — updated 2026-07-21
  Anthropic introduces Claude Fable 5, the state-of-the-art model for ambitious, long-running coding and knowledge work projects.
- [Claude Haiku 4.5: Near-Frontier Performance](https://howaiworks.ai/blog/claude-haiku-4-5-announcement.md) — updated 2025-10-15
  Anthropic announces Claude 4.5 Haiku, bringing frontier-level intelligence to its fastest and most affordable model tier with enhanced reasoning capabilities.
- [Claude Masterclass: The Essential Guide to AI Workflows](https://howaiworks.ai/blog/claude-masterclass-essential-workflows-2026.md) — updated 2026-04-21
  Unlock the full potential of Anthropic's Claude with this curated masterlist of tools, guides, and strategic frameworks for 2026.
- [Claude Opus 4.5: Best AI for Coding & Agents](https://howaiworks.ai/blog/claude-opus-4-5-announcement-2025.md) — updated 2025-11-25
  Anthropic releases Claude Opus 4.5 with state-of-the-art coding performance, improved efficiency, and new effort parameter for developers.
- [Claude Opus 5: Near-Fable Intelligence at Half the Price](https://howaiworks.ai/blog/claude-opus-5-announcement-2026.md) — updated 2026-07-24
  Anthropic released Claude Opus 5 on July 24, 2026 at $5/$25 per million tokens — thinking on by default, 26% fewer tokens than Opus 4.8.
- [Claude Sonnet 4.5: Anthropic's Advanced Model](https://howaiworks.ai/blog/claude-sonnet-4-5-announcement.md) — updated 2025-09-29
  Anthropic releases Claude Sonnet 4.5 with state-of-the-art coding capabilities, improved reasoning, and the new Claude Agent SDK for developers.
- [ClawdBot: The Open Source Personal AI Assistant](https://howaiworks.ai/blog/clawdbot-personal-ai-assistant.md) — updated 2026-01-26
  ClawdBot is an open-source, self-hosted AI agent that connects to your favorite chat apps and executes real-world tasks through your own infrastructure.
- [Cloudflare Announces Monetization Layer for the Agentic Web via x402](https://howaiworks.ai/blog/cloudflare-agentic-web-monetization-x402.md) — updated 2026-07-08
  Cloudflare introduces a new monetization layer allowing AI agents to pay websites directly at the HTTP request level using x402 on the edge.
- [Collaborator: A Unified macOS Canvas for Agentic Development](https://howaiworks.ai/blog/collaborator-macos-agent-workspace.md) — updated 2026-04-02
  Collaborator releases a dedicated macOS agent workspace, providing a seamless local environment for autonomous AI agents to interact with system tools.
- [Guide to Automotive Lidar Technology](https://howaiworks.ai/blog/comprehensive-automotive-lidar-guide-2025.md) — updated 2025-12-05
  Guide to automotive LiDAR in 2025, covering advancements in solid-state sensors, long-range detection, and AI-driven point cloud processing for safer driving.
- [Context Window vs Tokens vs Memory: What Limits an AI Chat](https://howaiworks.ai/blog/context-window-vs-tokens-vs-memory.md) — updated 2026-07-20
  Tokens, context window, and memory are three different things people mix up. Here's what each one is and which one actually limits your AI conversation.
- [Cursor 2.0: Composer, Multi-Agent UI, Terminal](https://howaiworks.ai/blog/cursor-2-0-composer-multi-agent-announcement.md) — updated 2025-10-29
  Cursor 2.0 launches Composer (fast agentic coding), parallel multi-agent workflows, GA browser testing, improved reviews, and sandboxed terminals.
- [Introducing Cursor 3: A Unified Agentic Workspace](https://howaiworks.ai/blog/cursor-3-unified-workspace-agents.md) — updated 2026-07-10
  Introducing Cursor 3, a workspace designed from the ground up for AI agents, featuring multi-repo support and seamless local-cloud handoff.
- [Cursor CLI: AI-Powered Terminal Assistant](https://howaiworks.ai/blog/cursor-cli-announcement.md) — updated 2025-09-28
  Cursor launches CLI tool bringing AI assistance directly to your terminal. Code faster with intelligent command-line automation and script generation.
- [Introducing Composer 2: The New Frontier of AI Coding in Cursor](https://howaiworks.ai/blog/cursor-composer-2-announcement.md) — updated 2026-07-08
  Cursor launches Composer 2, a next-gen model with record-breaking results on SWE-bench and Terminal-Bench, trained for long-horizon autonomous tasks.
- [Dario Amodei Proposes Exponential AI Policy](https://howaiworks.ai/blog/dario-amodei-ai-policy-exponential-2026.md) — updated 2026-06-12
  Anthropic CEO Dario Amodei outlines a 5-point policy framework to regulate AI development, addressing security, economy, and global leadership.
- [DeepSeek Plans a 160,000-Chip Huawei Cluster in Inner Mongolia](https://howaiworks.ai/blog/deepseek-huawei-ascend-160k-cluster-2026.md) — updated 2026-09-04
  Bloomberg reports DeepSeek will deploy at least 160,000 Huawei Ascend 950DT accelerators in Inner Mongolia — for inference. Training stays on Nvidia.
- [DeepSeek Teases Multimodal Capabilities: 'Now, We See You'](https://howaiworks.ai/blog/deepseek-multimodal-model-teaser.md) — updated 2026-05-01
  Xiaokang Chen of DeepSeek's multimodal team hints at upcoming vision features, signaling the lab's move toward integrated visual data understanding.
- [DeepSeek: Revolutionary OCR Context Compression](https://howaiworks.ai/blog/deepseek-ocr-context-compression-2025.md) — updated 2026-07-08
  DeepSeek researchers introduce a novel OCR context compression technique that reduces token usage by 80% while maintaining accuracy on complex document tasks.
- [DeepSeek-V3.2: GPT-5 Level Reasoning & Agent AI](https://howaiworks.ai/blog/deepseek-v3-2-release-2025.md) — updated 2025-12-01
  DeepSeek releases V3.2 and V3.2-Speciale models with GPT-5 level performance, gold-medal reasoning capabilities, and thinking in tool-use for agents.
- [DeepSeek-V4: Pro and Flash Models with 1M Context](https://howaiworks.ai/blog/deepseek-v4-pro-flash-release.md) — updated 2026-04-24
  DeepSeek releases V4-Pro and V4-Flash models featuring 1.6T parameters, open-source weights, and a massive 1 million token context window.
- [DeepSeek Makes Its 75% V4-Pro Discount Permanent](https://howaiworks.ai/blog/deepseek-v4-pro-price-reduction.md) — updated 2026-07-17
  DeepSeek's launch promo for V4-Pro was due to expire on May 31. Instead the company made it permanent, fixing the model at $0.435 and $0.87 per million tokens.
- [Doubao-Seed-Code: ByteDance's New Coding Model](https://howaiworks.ai/blog/doubao-seed-code-launch-2025.md) — updated 2025-12-03
  ByteDance's Volcano Engine launches Doubao-Seed-Code, achieving state-of-the-art on SWE-Bench-Verified with 62.7% lower costs and 256k context.
- [Dyson CameraJet: A $499 Toothbrush That Runs Computer Vision](https://howaiworks.ai/blog/dyson-camerajet-ai-toothbrush-2026.md) — updated 2026-09-04
  Dyson's CameraJet puts a 100,000-pixel camera in a toothbrush head and uses on-device machine vision to spot gaps between teeth and jet them clean.
- [ElevenLabs Music Marketplace: Monetize Your AI-Generated Tracks](https://howaiworks.ai/blog/elevenlabs-music-marketplace-announcement.md) — updated 2026-03-20
  ElevenLabs launches Music Marketplace in ElevenCreative, allowing users to publish, license, and earn from their AI-generated music tracks for commercial use.
- [Spatial Intelligence: AI's Next Frontier](https://howaiworks.ai/blog/fei-fei-li-spatial-intelligence-next-frontier-2025.md) — updated 2025-11-10
  Fei-Fei Li explains why spatial intelligence is AI's next breakthrough, enabling machines to understand and interact with the physical world.
- [Figure Launches Index, Paying People to Film Everyday Chores](https://howaiworks.ai/blog/figure-index-robot-training-dataset-2026.md) — updated 2026-08-26
  Figure's Index app has paid $15M to 264,000 users filming household and workplace tasks, building training data for its Helix robotics stack.
- [Garry Tan's Claude Code Senior Engineer Prompt](https://howaiworks.ai/blog/garry-tan-claude-code-senior-engineer-prompt.md) — updated 2026-02-23
  Discover how Y Combinator CEO Garry Tan uses Claude Code as a senior engineer to ship complex features with fully tested code in under an hour.
- [gstack: Garry Tan's Full Claude Code Setup, Explained](https://howaiworks.ai/blog/garry-tan-gstack-claude-code-setup.md) — updated 2026-07-12
  gstack is Garry Tan's open-source Claude Code setup — 23 opinionated skills that run Claude like an engineering team. What it does, and how to install it.
- [Gemini 3.1 Pro: A Smarter Model for Complex Tasks](https://howaiworks.ai/blog/gemini-3-1-pro-announcement.md) — updated 2026-02-23
  Google announces Gemini 3.1 Pro, featuring a 77.1% ARC-AGI-2 score and advanced reasoning for agentic workflows and complex system synthesis.
- [Gemini 3 Flash: Frontier Intelligence for Speed](https://howaiworks.ai/blog/gemini-3-flash-announcement-2025.md) — updated 2025-12-20
  Google releases Gemini 3 Flash, a fast AI model with Pro-grade reasoning at Flash-level speed. Achieves 90.4% on GPQA Diamond and 3x faster than 2.5 Pro.
- [Gemini 3 Rumors: Google's Next AI Model](https://howaiworks.ai/blog/gemini-3-rumors-2025.md) — updated 2025-11-17
  Early rumors about Google's Gemini 3 suggest a focus on real-time world simulation, integrated agentic reasoning, and a 100M+ token context window.
- [GEO vs SEO: What Changes When AI Answers the Query](https://howaiworks.ai/blog/geo-vs-seo-what-changes.md) — updated 2026-07-20
  AI answers resolve most searches without a click, and the pages they cite aren't the ones ranking #1. What GEO changes, and what carries over from SEO.
- [The Ralph Technique: Geoffrey Huntley's Agentic Coding Loop](https://howaiworks.ai/blog/geoffrey-huntley-ralph-agentic-coding-loop.md) — updated 2026-07-12
  The Ralph technique runs a coding agent in an infinite loop on one prompt. A plain guide to loop engineering, its $297 result, and the critique.
- [GitHub Copilot SDK: Build AI Agents Anywhere](https://howaiworks.ai/blog/github-copilot-sdk-agentic-apps.md) — updated 2026-01-26
  GitHub launches the Copilot SDK for building agentic applications, enabling developers to integrate Copilot's reasoning directly into their own software.
- [GLM-4.6: Zhipu AI's Advanced Coding Model](https://howaiworks.ai/blog/glm-4-6-announcement.md) — updated 2025-09-30
  Zhipu AI releases GLM-4.6 with 200K context window, enhanced coding capabilities, and 15% token efficiency improvements for real-world development tasks.
- [GLM-4.7-Flash: King of MoE Models in 30B Class](https://howaiworks.ai/blog/glm-4-7-flash-announcement.md) — updated 2026-01-21
  Z.ai releases GLM-4.7-Flash, a 30B MoE model with exceptional reasoning, coding, and agentic capabilities, rivaling much larger models.
- [GLM-5.3-Flash: 320B Multimodal MoE With a 1M-Token Context](https://howaiworks.ai/blog/glm-5-3-flash-announcement.md) — updated 2026-08-27
  Z.ai released GLM-5.3-Flash on August 26, 2026 — a 320B/18B-active multimodal MoE whose hybrid attention cuts KV cache 4.44x versus GLM-5.3.
- [GLM-5: Beyond Vibe Coding to Agentic Engineering](https://howaiworks.ai/blog/glm-5-announcement-2026.md) — updated 2026-02-16
  Zhipu AI unveils GLM-5, a state-of-the-art model designed for complex multi-file software engineering and long-horizon autonomous tasks.
- [GLM-5V-Turbo: The AI That Sees Your Screen and Writes the Code](https://howaiworks.ai/blog/glm-5v-turbo-vision-to-code.md) — updated 2026-04-02
  GLM-5V-Turbo is a native multimodal model that transforms designs, screenshots, and UI layouts into runnable code with unprecedented accuracy.
- [Google Launches Agent Payments Protocol (AP2)](https://howaiworks.ai/blog/google-agent-payments-protocol-ap2-announcement.md) — updated 2025-09-17
  Google announces AP2, an open protocol for secure AI agent payments with 60+ industry partners including Mastercard, PayPal, and Coinbase.
- [Google Antigravity Triples Gemini Request Limits](https://howaiworks.ai/blog/google-antigravity-triples-gemini-limits.md) — updated 2026-05-22
  Google Antigravity team member Varun Mohan announces a permanent 3x increase in Gemini request limits for paid tiers and resets weekly user quotas.
- [Google CEO Warns of AI Investment Irrationality](https://howaiworks.ai/blog/google-ceo-pichai-ai-investment-irrationality-warning-2025.md) — updated 2025-11-19
  Sundar Pichai warns of irrationality in AI investment cycles, comparing current boom to dotcom era while acknowledging no company is immune if bubble bursts.
- [Google Cloud Launches Advent of Agents 2025](https://howaiworks.ai/blog/google-cloud-advent-of-agents-2025.md) — updated 2025-12-03
  Google Cloud launches Advent of Agents 2025: a 25-day program to build production-ready AI agents using ADK, Agent Engine, and Gemini 3 models.
- [Google's Code Wiki: Accelerating Understanding](https://howaiworks.ai/blog/google-code-wiki-announcement-2025.md) — updated 2025-11-17
  Google launches Code Wiki, an AI-powered platform that automatically generates and maintains structured documentation for code repositories using Gemini.
- [Google Coral NPU: Full-Stack Platform for Edge AI](https://howaiworks.ai/blog/google-coral-npu-announcement-2025.md) — updated 2026-07-21
  Google announces Coral NPU, an open-source platform for ultra-low-power edge AI with RISC-V architecture, enabling all-day AI on wearables and IoT devices.
- [ATLAS: New Scaling Laws for Multilingual AI Models](https://howaiworks.ai/blog/google-deepmind-atlas-multilingual-scaling-laws.md) — updated 2026-01-29
  Google DeepMind's Atlas research explores multilingual scaling laws, providing a framework for training highly efficient models across hundreds of languages.
- [Google DeepMind Unveils Deep Research and Deep Research Max](https://howaiworks.ai/blog/google-deepmind-deep-research-announcement.md) — updated 2026-04-22
  Google DeepMind introduces autonomous research agents powered by Gemini 3.1 Pro, featuring native MCP support for private data analysis.
- [Google DESIGN.md: Standard for AI-Native Design Systems](https://howaiworks.ai/blog/google-design-md-standard-ai-agents.md) — updated 2026-04-24
  Google introduces DESIGN.md, an open-source standard bridging brand guidelines and AI coding agents for consistent, high-quality UI generation.
- [Google Earth AI: Geospatial Reasoning Updates](https://howaiworks.ai/blog/google-earth-ai-updates-2025.md) — updated 2025-10-24
  Google announces major updates to Earth AI, including Geospatial Reasoning powered by Gemini and expanded access to advanced geospatial data analysis.
- [Google Unveils Eighth-Generation TPUs: TPU 8t and TPU 8i](https://howaiworks.ai/blog/google-eighth-generation-tpu-8t-8i.md) — updated 2026-04-23
  Google introduces specialized TPU architectures for the agentic era, featuring TPU 8t for training and TPU 8i for inference and reasoning.
- [How a Google Engineer Automated 80% of Coding with Claude Code](https://howaiworks.ai/blog/google-engineer-automates-work-claude-code.md) — updated 2026-04-18
  Discover how a Google developer used Claude Code, a comprehensive CLAUDE.md file, and the Everything Claude Code OS to automate his routine tasks.
- [Google's Fairwind Program Gates Its Least-Restricted Cyber Model](https://howaiworks.ai/blog/google-fairwind-program-2026.md) — updated 2026-09-02
  Google launched Fairwind on September 2, 2026: vetted partners get Gemini 3.8 Flash Cyber — the variant that ships with deliberately looser safety mitigations.
- [Google Launches Gemini 2.5 Computer Use Model](https://howaiworks.ai/blog/google-gemini-2-5-computer-use-announcement.md) — updated 2025-10-08
  Google releases Gemini 2.5 Computer Use model, enabling AI agents to interact with user interfaces through web browsers and mobile apps with lower latency.
- [Google Updates Gemini 2.5 Flash Models](https://howaiworks.ai/blog/google-gemini-2-5-flash-flash-lite-update.md) — updated 2025-09-25
  Google releases improved Gemini 2.5 Flash and Flash-Lite models with 50% cost reduction, better agentic capabilities, and enhanced multimodal features.
- [Google Ships Gemini 3.6 Flash at a Lower Price Than 3.5](https://howaiworks.ai/blog/google-gemini-3-6-flash-launch-2026.md) — updated 2026-07-21
  Google released Gemini 3.6 Flash on July 21, 2026 — output priced at $7.50 per million tokens, a March 2026 knowledge cutoff, and 17% fewer output tokens.
- [Gemini 3.8 Flash: Same Price Per Token, 40% More Per Task](https://howaiworks.ai/blog/google-gemini-3-8-flash-launch-2026.md) — updated 2026-09-06
  Google's Gemini 3.8 Flash shipped September 2, 2026 at $0.75/$3.75 per million. Artificial Analysis measured cost per task rising 40% anyway.
- [Google Launches Gemini 3: Most Intelligent AI](https://howaiworks.ai/blog/google-gemini-3-announcement-2025.md) — updated 2025-11-18
  Google introduces Gemini 3, its most intelligent AI model with enhanced reasoning, multimodality, and coding capabilities, plus new Google Antigravity platform.
- [Gemini Now Picks Which Parts of a Video to Watch](https://howaiworks.ai/blog/google-gemini-agentic-video-2026.md) — updated 2026-09-02
  Google's agentic video understanding lets Gemini load video segments on demand. An hour of footage drops from ~475K tokens to a fraction of that.
- [Google Tests Gemini-Powered Conversational Search Ads](https://howaiworks.ai/blog/google-gemini-conversational-search-ads.md) — updated 2026-05-22
  Google is testing Gemini-powered conversational search ads, introducing interactive chatbots, dynamic product bundling, and direct native checkout.
- [Google Gemini Enterprise: AI for Workplace](https://howaiworks.ai/blog/google-gemini-enterprise-announcement.md) — updated 2025-10-11
  Google launches Gemini Enterprise, a comprehensive AI platform that unifies models, agents, and workflows to transform how organizations work, run.
- [Automate Your Workflows with Gemini Scheduled Actions](https://howaiworks.ai/blog/google-gemini-scheduled-actions-announcement.md) — updated 2026-03-03
  Learn how to use Gemini's new recurring actions to schedule daily summaries, weekly updates, and automated reports directly in the Gemini App.
- [Google Gemma 4: The Next Frontier of Open Models for AI Agents](https://howaiworks.ai/blog/google-gemma-4-release.md) — updated 2026-04-03
  Google introduces Gemma 4, a new family of open models optimized for complex reasoning, autonomous agents, and tool use with up to 256K context window.
- [Google Ironwood TPU and Axion VMs: AI Inference](https://howaiworks.ai/blog/google-ironwood-tpu-axion-vms-announcement-2025.md) — updated 2025-11-06
  Google announces Ironwood TPUs and Axion VMs, providing high-performance, energy-efficient infrastructure for training and deploying frontier AI models.
- [Google Labs Stitch: Future of AI-Native UI Design](https://howaiworks.ai/blog/google-labs-stitch-ai-ui-design.md) — updated 2026-03-19
  Google Labs unveils Stitch AI, a generative design tool that autonomously creates and iterates on UI components and design systems using simple prompts.
- [Google LAVA: AI-Powered VM Allocation](https://howaiworks.ai/blog/google-lava-vm-allocation-optimization.md) — updated 2025-10-19
  Google Research introduces LAVA, an AI-driven system that optimizes cloud computing resource allocation through advanced machine learning techniques.
- [LEAP: How LLMs Solved Every Putnam 2025 Problem](https://howaiworks.ai/blog/google-leap-putnam-2025.md) — updated 2026-06-09
  Google's LEAP paper details a system for automated theorem proving. It lets LLMs write formal Lean proofs and has solved 12 of 12 Putnam 2025 problems.
- [Google and Nvidia Shift Orders to Intel Amid TSMC Capacity Shortages](https://howaiworks.ai/blog/google-nvidia-intel-tsmc-shortage.md) — updated 2026-06-09
  Facing a shortage of TSMC production lines, Google and Nvidia are eyeing Intel as a backup chip maker. Google has already ordered 3 million TPUs for 2028.
- [Google Suncatcher: Space AI Infrastructure](https://howaiworks.ai/blog/google-project-suncatcher-space-based-ai-infrastructure.md) — updated 2025-11-04
  Google unveils Project Suncatcher, an ambitious initiative to build space-based AI infrastructure for global low-latency intelligence and sustainable compute.
- [Google's Science One Framework Ties AI Research Claims to Evidence](https://howaiworks.ai/blog/google-science-one-framework-chain-of-evidence.md) — updated 2026-08-02
  Google Research unveiled Science One, an autonomous research prototype whose Chain-of-Evidence design produced zero phantom citations versus baselines at 21%.
- [Google's SensorFM: A Foundation Model for Wearable Health Data](https://howaiworks.ai/blog/google-sensorfm-wearable-health-foundation-model-2026.md) — updated 2026-07-13
  Google Research introduces SensorFM, a foundation model trained on a trillion minutes of wearable sensor data from 5 million people to predict health outcomes.
- [Google Launches Google Skills: 3,000 AI Courses](https://howaiworks.ai/blog/google-skills-launch-2025.md) — updated 2026-07-08
  Google launches a new AI skills platform featuring certified courses and hands-on labs to help professionals master generative AI and machine learning.
- [Google Speculative Cascades: Faster LLM Inference](https://howaiworks.ai/blog/google-speculative-cascades-llm-inference.md) — updated 2025-09-15
  Google Research introduces speculative cascades, a revolutionary technique combining speculative decoding with cascades for better LLM inference.
- [Google TPUv7 Ironwood: Challenging Nvidia](https://howaiworks.ai/blog/google-tpuv7-ironwood-announcement-2025.md) — updated 2025-12-01
  Google's TPUv7 Ironwood commercializes AI chips externally. Anthropic's 1M TPU order signals potential end to Nvidia's CUDA dominance.
- [Google Workspace Oct: Veo 3.1, Security Updates](https://howaiworks.ai/blog/google-workspace-october-2025-updates.md) — updated 2025-11-21
  Google Workspace October 2025 updates: Veo 3.1 in Vids, Gemini in Sheets, ransomware protection, and expanded AI features across Workspace apps.
- [Google Workspace Studio: AI Agents for Work](https://howaiworks.ai/blog/google-workspace-studio-announcement-2025.md) — updated 2025-12-05
  Google launches Workspace Studio, enabling anyone to create AI agents that automate everyday work tasks using Gemini 3, with no coding required.
- [Introducing OpenAI GPT-5.4: New Frontier in AI Workflows](https://howaiworks.ai/blog/gpt-5-4-announcement-2026.md) — updated 2026-03-09
  OpenAI launches GPT-5.4 with 1M token context, native computer interaction, and 33% fewer errors. Discover how it redefines professional AI workflows.
- [Grok 4.1: xAI's Breakthrough in Emotional AI](https://howaiworks.ai/blog/grok-4-1-announcement.md) — updated 2025-11-18
  xAI releases Grok 4.1 with #1 LMArena ranking, 64.78% user preference, and enhanced creativity, emotional intelligence, and collaboration capabilities.
- [Grok-4 Fast: xAI's New Efficient AI Model](https://howaiworks.ai/blog/grok-4-fast-announcement.md) — updated 2025-09-20
  xAI announces Grok-4 Fast with 40% fewer tokens, 98% cost reduction, and 2M context window. Learn about the new efficient AI model.
- [Holo3: H Company's SOTA Foundation Model for Desktop Agents](https://howaiworks.ai/blog/h-company-holo3-desktop-agent.md) — updated 2026-04-02
  H Company unveils Holo3, a high-performance Mixture-of-Experts model family that sets a new industry standard for autonomous desktop application control.
- [Harper Reed's LLM Codegen Workflow, Explained](https://howaiworks.ai/blog/harper-reed-llm-codegen-workflow.md) — updated 2026-07-12
  Harper Reed's LLM codegen workflow is a three-stage spec, plan, execute method. See the exact prompts, the files it produces, and where it breaks down.
- [How to Tell If a Photo, Video or Text Is AI-Generated](https://howaiworks.ai/blog/how-ai-generated-content-is-detected.md) — updated 2026-07-13
  Detectors are unreliable and getting worse. How AI-content detection really works, which tools actually check a file, and the labeling laws now in force.
- [From Token to Transistor: What Happens When You Send a Prompt](https://howaiworks.ai/blog/how-ai-runs-on-silicon.md) — updated 2026-07-13
  Follow one prompt through the machine — tokens, matrix multiplications, the memory that everything waits on, and the chips built around a single operation.
- [How to Get Cited by ChatGPT and Perplexity](https://howaiworks.ai/blog/how-to-get-cited-by-chatgpt-perplexity.md) — updated 2026-07-20
  AI answer engines cite sources, and a citation now beats a blue link. Here's how to structure content so ChatGPT, Perplexity, and AI Overviews quote you.
- [HunyuanImage-3.0: Tencent's Massive 80B MoE Multimodal Model](https://howaiworks.ai/blog/hunyuan-image-3-0-announcement.md) — updated 2026-01-29
  Tencent releases HunyuanImage-3.0, the largest open-source MoE image generation model with a unified multimodal architecture and 80 billion parameters.
- [HunyuanImage 3.0-Instruct: Tencent's Massive Native Multimodal Leap](https://howaiworks.ai/blog/hunyuan-image-3-0-instruct-launch-2026.md) — updated 2026-01-29
  Tencent releases HunyuanImage 3.0-Instruct, the world's largest open-source image MoE model with 80B parameters, unifying understanding and generation.
- [IBM Releases Toucan: Largest Tool-Calling Dataset](https://howaiworks.ai/blog/ibm-toucan-tool-calling-dataset.md) — updated 2025-10-24
  IBM and University of Washington release Toucan, a groundbreaking dataset of 1.5 million real-world tool-calling scenarios designed to train better AI agents.
- [ICLR 2026: 21% of Peer Reviews Are AI-Generated](https://howaiworks.ai/blog/iclr-2026-ai-generated-peer-reviews-controversy.md) — updated 2025-11-28
  ICLR 2026 faces controversy as 21% of peer reviews were fully AI-generated, with over half showing AI use, raising academic integrity concerns.
- [AI Agents Take On a Challenge to Reproduce ICML 2026 Papers](https://howaiworks.ai/blog/icml-2026-agent-reproduction-challenge.md) — updated 2026-07-18
  A community challenge on Hugging Face has AI coding agents reproduce claims from every ICML 2026 paper, with $4,000 in GPU credits for the best runs.
- [Thinking Machines Releases Inkling, Its First Open-Weights Model](https://howaiworks.ai/blog/inkling-open-weights-model-thinking-machines.md) — updated 2026-07-18
  Thinking Machines Lab ships Inkling, a 975B-parameter multimodal MoE under Apache 2.0 — the leading U.S. open-weights model at release, per Artificial Analysis.
- [Inworld Ships Realtime TTS-2 Voice Model Family](https://howaiworks.ai/blog/inworld-realtime-tts-2-launch-2026.md) — updated 2026-09-04
  Inworld completes its Realtime TTS-2 rollout with a Flash variant and per-character pricing. Where the model actually ranks on the Artificial Analysis arena.
- [Jimeng, Dreamina & Seedance: ByteDance's AI Video, Explained](https://howaiworks.ai/blog/jimeng-dreamina-seedance-bytedance-ai-video.md) — updated 2026-07-12
  Jimeng, Dreamina, Seedance and Seedream, decoded: which is the app, which is the model, and how to use ByteDance's AI video outside China.
- [Kaggle & Google Release Free AI Agents Guide](https://howaiworks.ai/blog/kaggle-google-introduction-to-agents-whitepaper-2025.md) — updated 2025-11-11
  Kaggle and Google publish free 42-page whitepaper on AI agents covering architectures, training methods, and LangChain/LangGraph frameworks.
- [Kimi K2.6: Running 1 Trillion Parameters Locally](https://howaiworks.ai/blog/kimi-k2-6-dynamic-gguf-local-deployment.md) — updated 2026-04-23
  Unsloth releases Dynamic GGUF versions of Kimi K2.6, enabling the 1T parameter model to run on high-end local setups with speeds exceeding 40 tokens per second.
- [Kling AI Launches O1 Multimodal Video Generator](https://howaiworks.ai/blog/kling-ai-o1-multimodal-launch-2025.md) — updated 2025-12-02
  Kling AI launches O1 multimodal model for video and image generation, enabling integrated content creation with advanced AI capabilities.
- [Kling AI Video 2.6: Native Audio Generation](https://howaiworks.ai/blog/kling-ai-video-2-6-native-audio-2025.md) — updated 2025-12-05
  Kling AI Video 2.6 introduces native audio generation, enabling users to create cinematic videos with synchronized sound effects and background music.
- [LangChain Doubles Down on DeepAgents v0.2](https://howaiworks.ai/blog/langchain-doubling-down-on-deepagents.md) — updated 2025-10-30
  LangChain ships DeepAgents v0.2: plugin backends, offloading big tool outputs, conversation summarization, and safer recovery from interrupted tool calls.
- [LeWorldModel: Yann LeCun's End-to-End JEPA Breakthrough](https://howaiworks.ai/blog/le-world-model-jepa-architecture.md) — updated 2026-03-24
  Yann LeCun introduces LeWorldModel (LeWM), the first end-to-end JEPA trained from raw pixels, solving the collapse problem in world models.
- [LingBot-Depth: Precision Spatial Perception for Embodied AI](https://howaiworks.ai/blog/lingbot-depth-announcement.md) — updated 2026-02-02
  LingBot-Depth is a high-precision spatial perception model from Robbyant that delivers metrically accurate 3D measurements for robots and autonomous systems.
- [LinkedIn AI Cookbook: Scaling People Search](https://howaiworks.ai/blog/linkedin-generative-ai-cookbook-people-search-2025.md) — updated 2025-11-17
  LinkedIn reveals how it scaled generative AI-powered people search to 1.3 billion users using model distillation and collaborative design techniques.
- [Liquid AI LFM2.5-350M: A Sub-500MB Agentic Powerhouse](https://howaiworks.ai/blog/liquid-ai-lfm2-5-350m-agentic-model.md) — updated 2026-07-21
  Liquid AI releases LFM2.5-350M, a 350M parameter model trained on 28T tokens with RL, optimized for data extraction and agentic loops on edge devices.
- [Liquid AI LFM2.5-1.2B-Thinking: On-Device Reasoning Under 1GB](https://howaiworks.ai/blog/liquid-ai-lfm2-5-thinking-on-device.md) — updated 2026-07-21
  Liquid AI LFM2.5-1.2B-Thinking is a breakthrough 1.2B reasoning model that fits in 900MB, delivering high-performance logic on phones and laptops.
- [Liquid AI LFM2.5-1.2B-Thinking: Compact Power](https://howaiworks.ai/blog/liquidai-lfm2-5-1-2b-thinking-release.md) — updated 2026-07-21
  Exploring Liquid AI's newest 1.2B reasoning model optimized for agentic tasks, RAG, and high-speed edge inference with LIV convolution architecture.
- [LongCat-Flash-Omni: 560B Omni-Modal Model](https://howaiworks.ai/blog/longcat-flash-omni-announcement.md) — updated 2025-11-11
  LongCat releases Flash-Omni, a multi-modal reasoning model that achieves industry-leading performance on real-time vision and audio understanding benchmarks.
- [Lovable vs Bolt: Which AI App Builder Should You Use?](https://howaiworks.ai/blog/lovable-vs-bolt.md) — updated 2026-07-10
  Both turn prompts into full-stack apps. Bolt gives you a terminal and mobile output; Lovable gives you guardrails and a simpler stack. Here's how to choose.
- [Lovable vs v0: Which AI App Builder Should You Use?](https://howaiworks.ai/blog/lovable-vs-v0.md) — updated 2026-07-10
  v0 is Vercel's agentic builder built around Next.js. Lovable is self-contained and hides the stack entirely. The choice is mostly about where you already are.
- [Gemini App Lyria 3: Create Music from Text and Images](https://howaiworks.ai/blog/lyria-3-announcement.md) — updated 2026-02-19
  Create custom 30-second music tracks in Gemini using Google's Lyria 3 model. Generate songs from text or images with integrated SynthID watermarking.
- [Mac Mini M4: The Ultimate Hub for Autonomous AI Agents](https://howaiworks.ai/blog/mac-mini-m4-ai-agent-hub.md) — updated 2026-04-22
  Why the Mac mini M4 has become the 2026 industry standard for 24/7 AI agent hosting and local LLM pipelines.
- [Mafin 2.5: Reasoning RAG Hits 98.7% Accuracy](https://howaiworks.ai/blog/mafin-2-5-reasoning-rag-finance-breakthrough.md) — updated 2026-01-26
  Discover how Mafin 2.5 and the PageIndex framework are revolutionizing financial document analysis by replacing vector similarity with structured reasoning.
- [First Complete Male Fruit Fly Connectome Mapped With AI](https://howaiworks.ai/blog/male-fruit-fly-connectome-google-janelia-2026.md) — updated 2026-09-04
  Google Research, HHMI Janelia, MRC LMB and Cambridge published the first full wiring diagram of a male fruit fly nervous system: 166,691 neurons, 11,691 types.
- [Manus App Sharing: Streamlining Mobile Dev](https://howaiworks.ai/blog/manus-app-sharing-testing-publishing-2026.md) — updated 2026-01-21
  Manus simplifies the path from app development to real-world testing with automated AAB packaging for Android and direct TestFlight integration for iOS.
- [Computer Use Large: The Largest Open-Source Dataset for AI Agents](https://howaiworks.ai/blog/markov-ai-computer-use-large-dataset.md) — updated 2026-03-16
  Markov AI releases 'computer-use-large', a massive dataset of 48,000+ screen recordings for training AI agents to use professional software.
- [MCP Server Not Connecting? A Debug Checklist for 2026](https://howaiworks.ai/blog/mcp-server-not-connecting-fix.md) — updated 2026-07-20
  A step-by-step checklist for fixing MCP servers that fail to connect in Claude Code, Cursor, and Claude Desktop — from JSON errors to transport handshakes.
- [MCP vs A2A vs ACP: How AI Agents Actually Talk to Each Other](https://howaiworks.ai/blog/mcp-vs-a2a-vs-acp-how-ai-agents-talk.md) — updated 2026-07-20
  MCP connects agents to tools, A2A connects agents to each other, and ACP merged into A2A in 2025. A clear map of the agent protocol stack in 2026.
- [mem-agent: AI Model with Persistent Memory](https://howaiworks.ai/blog/mem-agent-persistent-memory-ai.md) — updated 2025-09-14
  mem-agent: A 4B parameter AI model with persistent memory that rivals models 50x larger. Trained with reinforcement learning on Obsidian-like memory systems.
- [Meta Develops Mango and Avocado AI Models](https://howaiworks.ai/blog/meta-mango-avocado-ai-announcement-2025.md) — updated 2025-12-20
  Meta announces Mango image/video AI model and Avocado LLM, targeting first-half 2026 release. Led by Scale AI founder Alexandr Wang with $14B investment.
- [Meta Launches Muse Spark 1.1, Its First Paid AI Model](https://howaiworks.ai/blog/meta-muse-spark-1-1-launch-2026.md) — updated 2026-07-13
  Meta released Muse Spark 1.1, a multimodal agentic model with a 1M-token context window, priced at $1.25/$4.25 per million tokens via a new public API.
- [Meta Ships Muse Spark 1.3 With Big Long-Context Gains](https://howaiworks.ai/blog/meta-muse-spark-1-3-release-2026.md) — updated 2026-09-04
  Muse Spark 1.3 posts 98.5 on MRCR and cuts tokens 25% on coding — but cost per task rose 37%, and the headline scores use a mode that has not shipped.
- [Meta Omnilingual ASR: 1,600+ Languages Support](https://howaiworks.ai/blog/meta-omnilingual-asr-announcement-2025.md) — updated 2025-11-10
  Meta introduces Omnilingual ASR supporting 1,600+ languages including 500 low-resource languages, with in-context learning for new languages.
- [Mark Zuckerberg's AI Assistant for CEO Tasks](https://howaiworks.ai/blog/meta-zuckerberg-ai-ceo-assistant.md) — updated 2026-03-24
  Mark Zuckerberg is building an AI agent to streamline his duties as Meta's CEO, enabling a flatter organization and reducing management layers.
- [Microsoft Adopts Claude Code from Anthropic](https://howaiworks.ai/blog/microsoft-claude-code-adoption.md) — updated 2026-01-26
  Microsoft is deploying Anthropic's Claude Code for its internal teams, favoring it over GitHub Copilot in key development scenarios.
- [Microsoft MAI-Image-1: Top 10 Image Gen Model](https://howaiworks.ai/blog/microsoft-mai-image-1-announcement.md) — updated 2025-10-14
  Microsoft releases MAI-Image-1, a next-generation diffusion model featuring state-of-the-art text-to-image consistency and advanced architectural efficiency.
- [Microsoft MAI-Image-2: A New Frontier for Photorealistic AI Imagery](https://howaiworks.ai/blog/microsoft-mai-image-2-announcement.md) — updated 2026-03-20
  Microsoft unveils MAI-Image-2: a creator-focused successor with enhanced photorealism, reliable in-image text, and hyper-detailed scene generation.
- [MiniMax H3: 2K Video With Native Audio, Weights Shipped](https://howaiworks.ai/blog/minimax-h3-open-weights-video-model.md) — updated 2026-09-06
  MiniMax announced H3 on July 31, 2026: one model taking text, images, video and audio, out to 2K video with stereo sound. The weights shipped August 3.
- [MiniMax M2.7: Early Echoes of AI Self-Evolution](https://howaiworks.ai/blog/minimax-m27-self-evolution.md) — updated 2026-03-19
  MiniMax unveils M2.7, a breakthrough model that participates in its own evolution through autonomous agent harnesses and advanced software engineering.
- [MiniMax M3 Open-Sourced on Hugging Face](https://howaiworks.ai/blog/minimax-m3-open-source-hugging-face.md) — updated 2026-06-12
  MiniMax has open-sourced its M3 model, a 428B MoE architecture optimized for long context and agentic scenarios, now available on Hugging Face.
- [Mistral 3: Next Generation Open Multimodal AI](https://howaiworks.ai/blog/mistral-3-announcement-2025.md) — updated 2026-07-21
  Mistral AI announces Mistral 3 with Large 3 and Ministral 3 series, featuring state-of-the-art performance, multimodal capabilities, and Apache 2.0 licensing.
- [MIT Deep Learning Fall 2024 Course Released for Free](https://howaiworks.ai/blog/mit-deep-learning-fall-2024-course-release.md) — updated 2026-04-21
  MIT OpenCourseWare has released the full '6.7960 Deep Learning' course from Fall 2024, featuring Phillip Isola and comprehensive materials for self-study.
- [Embedded Language Flows: MIT Revitalizes Text Diffusion](https://howaiworks.ai/blog/mit-elf-embedded-language-flows.md) — updated 2026-05-21
  MIT researchers introduce Embedded Language Flows (ELF), a continuous-time flow matching framework bringing data-efficient diffusion models to text generation.
- [Mixture of Experts Explained: Why Every 2026 Model Uses It](https://howaiworks.ai/blog/mixture-of-experts-explained.md) — updated 2026-07-20
  Mixture of Experts lets a model have trillions of parameters but only use a fraction per token. Here's how it works and why it took over in 2026.
- [MobileLLM-Pro: Meta's 1B On-Device Model](https://howaiworks.ai/blog/mobilellm-pro-announcement.md) — updated 2026-07-21
  Meta Reality Labs releases MobileLLM-Pro, a 1B parameter language model optimized for on-device inference with 128k context window and near-lossless int4.
- [Introducing Our New AI Models Section](https://howaiworks.ai/blog/models-section-launch.md) — updated 2026-07-08
  We've added a comprehensive /models section with detailed profiles of frontier AI models from OpenAI, Anthropic, Google, Meta, and leading open-weight labs.
- [Kimi K2.6 Release: Open Weights and 12-Hour Long-Horizon Coding](https://howaiworks.ai/blog/moonshot-kimi-k2-6-release-announcement.md) — updated 2026-07-08
  Moonshot AI releases Kimi K2.6, featuring open weights, impressive coding benchmarks, and support for agentic swarms with up to 300 sub-agents.
- [Moonshot Ships Kimi K3: 2.8T Parameters, 1M Context, Weights July 27](https://howaiworks.ai/blog/moonshot-kimi-k3-release-announcement.md) — updated 2026-07-17
  Moonshot AI's Kimi K3 replaces K2.6 as flagship with a new attention architecture, always-on thinking, and an open-weights promise dated July 27, 2026.
- [MWS AI Launches Cotype Light 3: 9B Multimodal](https://howaiworks.ai/blog/mws-ai-cotype-light-3-announcement.md) — updated 2026-04-03
  MWS AI launches Cotype-Light-3, an ultra-efficient 3B parameter model optimized for mobile and edge devices with near-lossless performance on core benchmarks.
- [Nanbeige4.1-3B: Compact Powerhouse with Strong Reasoning](https://howaiworks.ai/blog/nanbeige-4-1-3b-announcement.md) — updated 2026-02-16
  Nanbeige releases a 4.1.3B parameter model that sets new open-weights benchmarks for efficiency and reasoning on small-scale hardware and mobile devices.
- [Nano Banana: Google's AI Photo Editing Tool 2025](https://howaiworks.ai/blog/nano-banana-ai-image-editor.md) — updated 2025-09-10
  Nano-Banana launches an AI-powered image editor that uses diffusion models to perform complex edits, restyling, and object removal through simple text prompts.
- [NeurIPS 2025 Best Paper Awards: 7 Papers](https://howaiworks.ai/blog/neurips-2025-best-paper-awards-announcement.md) — updated 2025-11-26
  NeurIPS 2025 announces best paper awards, highlighting breakthroughs in Sparse Attention, agentic reasoning, and energy-efficient neural network architectures.
- [Qwen3 Max Leads NOF1 AI Arena: 79% Return](https://howaiworks.ai/blog/nof1-ai-arena-leaderboard-qwen3-max-leads.md) — updated 2025-10-26
  The latest NoF1 AI Arena leaderboard reveals Qwen3-Max as the top-performing model in reasoning and coding, surpassing global competitors in head-to-head tests.
- [NVIDIA's $12.9B Hugging Face Deal Is a Bet on Open Models](https://howaiworks.ai/blog/nvidia-acquires-hugging-face-open-source.md) — updated 2026-09-04
  NVIDIA will acquire Hugging Face for $12.9 billion. Why the deal's incentives favor open models — and where the neutrality risk actually sits.
- [NVIDIA Blackwell: Performance Leaps for MoE](https://howaiworks.ai/blog/nvidia-blackwell-moe-inference.md) — updated 2026-01-21
  Discover how NVIDIA Blackwell and TensorRT-LLM deliver up to 2.8x throughput increases for Mixture of Experts (MoE) models like DeepSeek-R1.
- [NVIDIA DGX Spark: Petaflop System for SpaceX](https://howaiworks.ai/blog/nvidia-dgx-spark-announcement.md) — updated 2025-10-14
  NVIDIA launches DGX Spark, the world's smallest AI supercomputer with 1 petaflop performance and 128GB memory, first delivered to Elon Musk at SpaceX.
- [NVIDIA Earth-2: World's First Open AI Weather Models](https://howaiworks.ai/blog/nvidia-earth-2-open-models.md) — updated 2026-01-29
  NVIDIA launches Earth-2, a family of open, accelerated models for global weather forecasting, nowcasting, and data assimilation.
- [NVIDIA CEO: Automate Every Task with AI](https://howaiworks.ai/blog/nvidia-jensen-huang-automate-every-task-2025.md) — updated 2025-11-29
  Jensen Huang tells NVIDIA employees to automate every possible task with AI, addressing concerns about job security as company grows to 36,000 workers.
- [NVIDIA Kimodo: AI-Powered 3D Motion Generation](https://howaiworks.ai/blog/nvidia-kimodo-3d-motion-generation.md) — updated 2026-03-24
  NVIDIA releases Kimodo, a diffusion-based generative model for realistic 3D motion, supporting diverse skeletons including SMPL-X and Unitree G1.
- [Nemotron 3.5 Lightning: One GPU, Yes. A Laptop, No.](https://howaiworks.ai/blog/nvidia-nemotron-3-5-lightning-local-gpu-reality.md) — updated 2026-09-06
  NVIDIA's 30B Nemotron 3.5 Lightning runs on a single desktop GPU, but its Q4_K_M GGUF is 25.3 GB — above every 24 GB mobile card.
- [NVIDIA Open-Sources Nemotron 3 Embed, Claiming the RTEB Top Spot](https://howaiworks.ai/blog/nvidia-nemotron-3-embed-rteb.md) — updated 2026-07-21
  NVIDIA open-sources Nemotron 3 Embed: the 8B model ranks #1 in NVIDIA's RTEB retrieval tests, and an NVFP4 build doubles throughput on Blackwell.
- [NVIDIA Omni-Embed-Nemotron-3B: Multimodal RAG](https://howaiworks.ai/blog/nvidia-omni-embed-nemotron-3b-announcement.md) — updated 2025-10-17
  NVIDIA releases Omni-Embed-Nemotron-3B, a versatile multimodal embedding model for text, image, audio, and video content in RAG systems.
- [How Open Models Are Driving AI Research: Insights from ICML 2026](https://howaiworks.ai/blog/nvidia-open-models-icml-2026.md) — updated 2026-07-06
  Explore how NVIDIA's open models like Nemotron, Cosmos, and BioNeMo are fueling major research breakthroughs at ICML 2026.
- [NVIDIA PersonaPlex: Controlled Full-Duplex Speech AI](https://howaiworks.ai/blog/nvidia-personaplex-announcement.md) — updated 2026-02-03
  Discover PersonaPlex, NVIDIA's breakthrough in full-duplex speech AI that allows precise control over voice and persona for natural, low-latency interactions.
- [NVIDIA RLP: RL Pretraining for AI Models](https://howaiworks.ai/blog/nvidia-rlp-reinforcement-learning-pretraining.md) — updated 2025-10-02
  NVIDIA introduces RLP (Reinforcement Learning Pretraining), a novel method for training base models with RL to enhance reasoning and strategic decision-making.
- [NVIDIA TTT-E2E: Test-Time Training Long Context](https://howaiworks.ai/blog/nvidia-ttt-e2e-test-time-training-long-context.md) — updated 2026-01-09
  NVIDIA researchers introduce TTT-E2E, a test-time training approach that enables models to handle ultra-long contexts efficiently through dynamic adaptation.
- [NVIDIA Claims 30x Efficiency Gain for AI Agents on Vera Rubin](https://howaiworks.ai/blog/nvidia-vera-rubin-nvl72-agentic-efficiency.md) — updated 2026-08-25
  NVIDIA reports Vera Rubin NVL72 delivers up to 30x higher throughput per megawatt and 35x lower token cost than GB300 NVL72 on agentic coding workloads.
- [Odyssey-2 Max: A New SOTA in Real-Time Physics World Models](https://howaiworks.ai/blog/odyssey-2-max-world-model-physics.md) — updated 2026-04-23
  Odyssey releases Odyssey-2 Max, an autoregressive world model achieving SOTA results in physics simulation and real-time user interactivity.
- [OpenAI Reaches $500B Valuation: World's Largest](https://howaiworks.ai/blog/openai-500-billion-valuation-2025.md) — updated 2025-10-02
  OpenAI completes $6.6B share sale at $500 billion valuation, surpassing SpaceX to become the world's most valuable startup.
- [OpenAI Accelerates Biological Research with GPT-5](https://howaiworks.ai/blog/openai-accelerating-biological-research-wet-lab-2025.md) — updated 2025-12-18
  OpenAI and Red Queen Bio use GPT-5 to optimize molecular cloning protocols, achieving 79-fold efficiency increase in wet lab research.
- [OpenAI Introduces AgentKit: AI Agent Platform](https://howaiworks.ai/blog/openai-agentkit-announcement.md) — updated 2025-10-06
  OpenAI launches AgentKit, a complete toolkit for building, deploying, and optimizing AI agents with visual builder, chat interfaces, and evaluation tools.
- [OpenAI & Broadcom: Strategic AI Partnership](https://howaiworks.ai/blog/openai-broadcom-strategic-collaboration.md) — updated 2025-10-13
  OpenAI partners with Broadcom to develop and deploy 10 gigawatts of custom AI accelerators, revolutionizing AI infrastructure for the next generation.
- [OpenAI Introduces Apps in ChatGPT: New Era](https://howaiworks.ai/blog/openai-chatgpt-apps-announcement.md) — updated 2025-10-06
  OpenAI launches ChatGPT apps integration with Booking.com, Spotify, Canva and more, plus Apps SDK for developers to build custom applications.
- [OpenAI to Transform ChatGPT into a Superapp with Autonomous Agents](https://howaiworks.ai/blog/openai-chatgpt-superapp-agents.md) — updated 2026-06-09
  OpenAI will soon roll out the first major redesign of ChatGPT since 2022, transforming it from a conversational chatbot into a platform for autonomous agents.
- [OpenAI Unveils Free Codex Skills: Transforming Assistants into Agents](https://howaiworks.ai/blog/openai-codex-free-skills-library-launch.md) — updated 2026-04-18
  OpenAI has just released a library of free, one-click skills for Codex, enabling robust automation for design, mobile development, and presentation workflows.
- [OpenAI Company Knowledge: Business Data Insights](https://howaiworks.ai/blog/openai-company-knowledge-chatgpt-enterprise.md) — updated 2025-10-23
  OpenAI launches Company Knowledge for ChatGPT Enterprise, enabling seamless integration with workplace tools like Slack and SharePoint for team productivity.
- [OpenAI Releases GPT-5.5: The Agentic Coding Revolution](https://howaiworks.ai/blog/openai-gpt-5-5-announcement.md) — updated 2026-04-24
  OpenAI has announced GPT-5.5 (Spud), a massive new base model optimized for agentic coding, handling complex 20-hour tasks with ease. Here are the details.
- [GPT-5.6 Ships as Three Models That Differ Only in Price](https://howaiworks.ai/blog/openai-gpt-5-6-sol-terra-luna-launch-2026.md) — updated 2026-07-21
  OpenAI's Sol, Terra and Luna went generally available July 9, 2026. All three share one context window, one output cap and one knowledge cutoff.
- [OpenAI GPT-5: Accelerating Scientific Research](https://howaiworks.ai/blog/openai-gpt-5-accelerating-science-2025.md) — updated 2025-11-21
  OpenAI reveals GPT-5 early experiments showing breakthrough capabilities in mathematics, physics, biology, and computer science research acceleration.
- [GPT-6 Astra Has No Realtime API and a 272K Pricing Cliff](https://howaiworks.ai/blog/openai-gpt-6-astra-api-migration-2026.md) — updated 2026-09-06
  Three days after launch, the GPT-6 Astra facts that decide a migration: no Realtime endpoint, a 272K billing threshold, and a quiet GPT-5.6 price cut.
- [OpenAI Launches GPT-6 Astra With Record Benchmarks](https://howaiworks.ai/blog/openai-gpt-6-astra-launch-2026.md) — updated 2026-09-04
  OpenAI announced GPT-6 Astra on September 3, 2026, claiming 97.6% on FrontierMath Tier 4 v2 and 100% on ExploitBench. The ARC-AGI-3 result comes with a caveat.
- [OpenAI Instant Checkout: Revolutionizing Shopping](https://howaiworks.ai/blog/openai-instant-checkout-announcement.md) — updated 2025-09-29
  OpenAI launches Instant Checkout in ChatGPT, revolutionizing AI commerce with seamless shopping directly in conversations using the new Agentic Commerce.
- [OpenAI Takes Its First Official Step Towards an IPO](https://howaiworks.ai/blog/openai-ipo-first-step-wsj.md) — updated 2026-06-09
  OpenAI has confidentially filed an S-1 form to go public, beginning the SEC review process.
- [OpenAI Parental Controls: AI Safety for Families](https://howaiworks.ai/blog/openai-parental-controls-2025.md) — updated 2025-09-29
  OpenAI introduces parental controls for ChatGPT, enabling safe AI usage for teens with account linking, content filtering, and parental oversight features.
- [OpenAI: Interpretability via Sparse Circuits](https://howaiworks.ai/blog/openai-sparse-circuits-neural-networks.md) — updated 2025-11-17
  OpenAI researchers publish findings on Sparse Circuits, revealing how neural networks process information through dedicated, interpretable internal pathways.
- [OpenClaw 2.0: Shared Sessions and Auto-Detected Credentials](https://howaiworks.ai/blog/openclaw-2-0-release.md) — updated 2026-09-01
  OpenClaw 2.0 ships simpler setup, automatic detection of ChatGPT and Claude sign-ins, a rebuilt web app, and cloud-backed shared sessions.
- [OpenClaw: 869 AI Skills for Medical Research](https://howaiworks.ai/blog/openclaw-medical-skills-ai-medicine-library.md) — updated 2026-07-08
  OpenClaw launches a medical skills library, providing open-source tools and datasets for AI agents to assist in clinical research and medical documentation.
- [OpenEnv: Standard Agent Training Environments](https://howaiworks.ai/blog/openenv-agentic-execution-environments.md) — updated 2025-10-26
  OpenEnv launches a standard for agentic execution environments, enabling AI agents to operate securely across cloud and local platforms with unified protocols.
- [OpenHarness: The Open-Source Infrastructure Layer for AI Agents](https://howaiworks.ai/blog/openharness-open-source-agent-infrastructure.md) — updated 2026-07-08
  OpenHarness launches an open-source infrastructure layer for AI agents, providing unified tools for memory management, tool validation, and secure execution.
- [Oracle AI Database 26ai: AI in Data Management](https://howaiworks.ai/blog/oracle-ai-database-26ai-announcement.md) — updated 2025-10-17
  Oracle announces AI Database 26ai, featuring native vector search and autonomous tuning to accelerate generative AI application development and scaling.
- [Oracle & OpenAI: $40B Partnership](https://howaiworks.ai/blog/oracle-openai-partnership-2025.md) — updated 2025-09-10
  Oracle and OpenAI's $40B partnership includes Nvidia chips and 4.5GW data centers. Learn about Stargate project and AI infrastructure impact.
- [PaddleOCR-VL-1.5: SOTA Multimodal Document Parsing](https://howaiworks.ai/blog/paddleocr-vl-1-5-announcement.md) — updated 2026-07-21
  Baidu announces PaddleOCR-VL-1.5, a 0.9B VLM achieving 94.5% on OmniDocBench v1.5 with breakthrough robustness in real-world scenarios.
- [PaddleOCR-VL: Baidu's 0.9B Vision-Language Model](https://howaiworks.ai/blog/paddleocr-vl-announcement.md) — updated 2025-10-19
  Baidu releases PaddleOCR-VL, a multi-modal OCR model that combines advanced visual perception with text recognition for superior document understanding.
- [Stanford Launches AI Agentic Paper Reviewer](https://howaiworks.ai/blog/paperreview-ai-stanford-agentic-reviewer-2025.md) — updated 2025-11-28
  Stanford ML Group releases PaperReview.ai, an agentic system that provides rapid research paper feedback grounded in latest arXiv publications.
- [Perplexity BrowseSafe: Safer AI Browsers](https://howaiworks.ai/blog/perplexity-browsesafe-announcement-2025.md) — updated 2025-12-05
  Perplexity introduces BrowseSafe, an open detection model and benchmark for protecting AI agents from prompt injection attacks in browser environments.
- [PyTorch Monarch: Distributed Programming](https://howaiworks.ai/blog/pytorch-monarch-announcement-2025.md) — updated 2026-07-08
  PyTorch announces Project Monarch, a new compiler backend that significantly improves training efficiency for Mixture-of-Experts (MoE) models on NVIDIA GPUs.
- [Qwen Code v0.2.2: Stability Improvements](https://howaiworks.ai/blog/qwen-code-v0-2-2-release-2025.md) — updated 2025-11-17
  Alibaba's Qwen Code CLI tool releases v0.2.2 with performance enhancements and bug fixes, continuing rapid development of this AI-powered developer tool.
- [Qwen's E-Commerce Bench Runs AI Agents for a Simulated Year](https://howaiworks.ai/blog/qwen-e-commerce-bench-2026.md) — updated 2026-09-04
  Qwen released E-Commerce Bench: an LLM agent gets ¥100,000 and 365 simulated days to run online stores. 18 frontier models tested, none wins overall.
- [Qwen3-ASR: SOTA Multilingual Speech Recognition and Forced Alignment](https://howaiworks.ai/blog/qwen3-asr-announcement.md) — updated 2026-01-30
  Alibaba's Qwen team releases Qwen3-ASR and Qwen3-ForcedAligner, setting new benchmarks in multilingual speech-to-text and precise timestamping.
- [Qwen3-Max-Thinking: A New Era for Reasoning Models](https://howaiworks.ai/blog/qwen3-max-thinking-announcement-2026.md) — updated 2026-01-27
  Alibaba Cloud introduces Qwen3-Max-Thinking, a flagship reasoning model with adaptive tool-use and test-time scaling, rivaling GPT-5.2 and Claude Opus 4.5.
- [Qwen3-Max-Thinking: Perfect Reasoning Scores](https://howaiworks.ai/blog/qwen3-max-thinking-perfect-scores-announcement.md) — updated 2025-11-05
  Alibaba's Qwen3-Max-Thinking achieves 100% on AIME 2025 and HMMT, matching OpenAI's top model on reasoning benchmarks while emphasizing step-by-step solutions.
- [Qwen3-TTS Open Sourced: Voice Design and Clone](https://howaiworks.ai/blog/qwen3-tts-open-source-announcement-2026.md) — updated 2026-01-23
  Alibaba open-sources Qwen3-TTS family with voice design, cloning, and ultra-high-quality speech generation across 10 languages.
- [Qwen3-VL Cookbooks: Guide to Multimodal Vision AI](https://howaiworks.ai/blog/qwen3-vl-cookbooks-guide.md) — updated 2025-10-12
  Explore Qwen3-VL Cookbooks with practical examples for multimodal AI development, including vision-language tasks, image analysis, and integration guides.
- [RND1: Largest Open Diffusion Language Model](https://howaiworks.ai/blog/radical-numerics-rnd1-diffusion-model.md) — updated 2025-10-12
  Radical Numerics introduces RND1-Base, a 30B parameter diffusion language model converted from autoregressive architecture with 15% efficiency gains.
- [Real-Qwen-Image-V2: New Era of AI Realism](https://howaiworks.ai/blog/real-qwen-image-v2-announcement-2026.md) — updated 2026-01-26
  Reviewing Real-Qwen-Image-V2 — a fine-tuned version of Qwen-Image-2512 focused on photorealism, sharpness, and optimized facial aesthetics.
- [Complete Roadmap of Math for Machine Learning](https://howaiworks.ai/blog/roadmap-mathematics-machine-learning-2025.md) — updated 2025-11-28
  A comprehensive guide to the three pillars of ML mathematics: linear algebra, calculus, and probability theory.
- [A Business Book as Slash Commands: Sahil Lavingia's Claude Skills](https://howaiworks.ai/blog/sahil-lavingia-minimalist-entrepreneur-claude-skills.md) — updated 2026-07-13
  Sahil Lavingia turned The Minimalist Entrepreneur into 10 Claude Code skills. What is inside the 9.5K-star repo, how to install it, and what it cannot do.
- [Sakana AI to Focus on Algorithmic Evolution of AI](https://howaiworks.ai/blog/sakana-ai-rsi-lab-announcement.md) — updated 2026-06-09
  The Japanese startup has opened a Recursive Self-Improvement (RSI) research lab to create networks that optimize their own code.
- [Sakana AI's Smart Cellular Bricks Recognize Their Own Shape](https://howaiworks.ai/blog/sakana-ai-smart-cellular-bricks-2026.md) — updated 2026-07-21
  Sakana AI, ITU Copenhagen and Autodesk built ~200 cubic bricks that figure out what shape they form — with no central controller and no position data.
- [Sakana AI Launches Sudoku-Bench for AI Reasoning](https://howaiworks.ai/blog/sakana-ai-sudoku-bench-announcement-2025.md) — updated 2025-11-11
  Sakana AI introduces Sudoku-Bench, a creative reasoning benchmark testing human-like problem-solving through Sudoku variants without tool use.
- [Sam Altman Pushes for OpenAI IPO in September](https://howaiworks.ai/blog/sam-altman-openai-ipo-september.md) — updated 2026-05-22
  OpenAI targets a September IPO as Sam Altman accelerates listing plans despite CFO cautions, following the dismissal of Elon Musk's lawsuit.
- [Seedance 2.0: Breakthroughs and Copyright Launch Delay](https://howaiworks.ai/blog/seedance-2-0-launch-and-copyright-controversy.md) — updated 2026-02-25
  ByteDance unveils Seedance 2.0, a powerful AI video model, but postpones its global launch amid significant copyright infringement allegations.
- [Seedance 2.5 Ships: 30-Second Video in a Single Pass](https://howaiworks.ai/blog/seedance-2-5-launch-2026.md) — updated 2026-08-02
  ByteDance Seed announced Seedance 2.5 on July 31, 2026: 30-second audio-video clips in one pass and up to 50 reference inputs per generation.
- [Self-Hosted vs API: When Running Your Own LLM Pays Off](https://howaiworks.ai/blog/self-hosted-vs-api-llm-cost.md) — updated 2026-07-20
  The raw GPU price is a trap. A 2026 breakdown of the real cost of self-hosting an LLM and where the break-even against API pricing actually sits.
- [Skill Seeker: Docs to Claude AI Skills](https://howaiworks.ai/blog/skill-seeker-transform-docs-into-claude-skills.md) — updated 2025-12-15
  Skill Seeker is a powerful open-source tool that converts documentation, GitHub repositories, and PDFs into optimized skills for Claude AI.
- [Sony AI Ace: First Robot to Beat Pro Table Tennis Players](https://howaiworks.ai/blog/sony-ai-ace-table-tennis-robot.md) — updated 2026-04-24
  Sony AI introduces Ace, a groundbreaking table tennis robot featuring a 20.2ms reaction time and advanced reinforcement learning to defeat human experts.
- [Sora 2: Advanced Video and Audio Generator](https://howaiworks.ai/blog/sora-2-announcement.md) — updated 2025-10-01
  OpenAI launches Sora 2, a state-of-the-art video and audio generation model with improved physics accuracy, synchronized audio, and enhanced safety features.
- [Step Audio R1: First Audio Reasoning Model](https://howaiworks.ai/blog/step-audio-r1-first-audio-reasoning-model-2025.md) — updated 2025-12-03
  Step Audio R1 is the first audio model to unlock Chain-of-Thought reasoning, solving inverted scaling and surpassing Gemini 2.5 Pro in complex audio tasks.
- [Step3-VL-10B: Redefining Multimodal AI](https://howaiworks.ai/blog/step3-vl-10b-announcement-2026.md) — updated 2026-01-26
  Stepfun AI releases Step3-VL-10B, a 10B parameter multimodal model that outperforms giants 20x its size through innovative Parallel Coordinated Reasoning.
- [T5Gemma 2: Next-Gen Encoder-Decoder Models](https://howaiworks.ai/blog/t5gemma-2-announcement-2025.md) — updated 2025-12-20
  Google DeepMind releases T5Gemma-2, a hybrid model combining the strengths of T5 and Gemma for industry-leading performance on translation and reasoning tasks.
- [Tencent HPC-Ops: SOTA Performance for LLM Inference](https://howaiworks.ai/blog/tencent-hpc-ops-announcement.md) — updated 2026-07-21
  Tencent releases HPC-Ops, a production-grade high-performance operator library for LLM inference, delivering up to 2.22x speedup on NVIDIA H20 GPUs.
- [Tencent Hunyuan 3D 3.0: Triple Modeling Accuracy](https://howaiworks.ai/blog/tencent-hunyuan-3d-3-0-announcement.md) — updated 2025-09-16
  Tencent releases Hunyuan-3D 3.0, a state-of-the-art generative model for high-fidelity 3D asset creation from text prompts and single images in seconds.
- [Tencent Hunyuan's UniRL: Universal RL for Multimodal Models](https://howaiworks.ai/blog/tencent-hunyuan-unirl-announcement.md) — updated 2026-06-09
  Tencent Hunyuan releases UniRL, a unified infrastructure for reinforcement learning post-training across LLMs, VLMs, and diffusion models.
- [Tencent HunyuanVideo-1.5: Efficient Video Gen](https://howaiworks.ai/blog/tencent-hunyuanvideo-1-5-announcement-2025.md) — updated 2025-11-21
  Tencent releases HunyuanVideo-1.5, a compact 8.3B-parameter video generation model with SSTA attention and 1080p super-resolution support.
- [Tencent HY 2.0: MoE Model with 73.4 IMO Score](https://howaiworks.ai/blog/tencent-hy-2-0-release-2025.md) — updated 2025-12-06
  Tencent releases HY 2.0 foundation model with MoE architecture, 256K context, and major gains in reasoning, coding, and instruction following capabilities.
- [Tencent HY-WU: Dynamic LoRA for Precise Image Editing](https://howaiworks.ai/blog/tencent-hy-wu-announcement.md) — updated 2026-03-09
  Tencent introduces HY-WU, a Weight Unleashing framework that generates dynamic LoRA adapters to solve gradient conflicts in multi-task image editing.
- [Terence Tao: The Future of AI in Mathematics](https://howaiworks.ai/blog/terence-tao-dwarkesh-patel-interview.md) — updated 2026-03-24
  Renowned mathematician Terence Tao discusses how AI is transforming mathematical research, idea generation, and the shift from creation to filtration.
- [How to Build an Agent: Thorsten Ball's 400-Line Blueprint](https://howaiworks.ai/blog/thorsten-ball-how-to-build-an-agent.md) — updated 2026-07-12
  Thorsten Ball's 'How to Build an Agent' shows a code-editing AI is just an LLM in a loop with three tools, in under 400 lines of Go.
- [TIGER-AI-Lab Open-Sources OpenResearcher Agent Pipeline](https://howaiworks.ai/blog/tiger-ai-lab-openresearcher-open-source.md) — updated 2026-07-20
  TIGER-AI-Lab released OpenResearcher, a fully open pipeline for training deep research agents, with 96K web-research trajectories and 54.8% on BrowseComp-Plus.
- [TPUs vs GPUs vs ASICs: AI Hardware Guide 2025](https://howaiworks.ai/blog/tpu-gpu-asic-ai-hardware-market-2025.md) — updated 2025-12-05
  Complete guide to TPUs, GPUs, and ASICs for AI workloads. Compare architectures, performance, efficiency, and market trends as of December 2025.
- [Training vs Inference: Two Economies, Two Opposite Bottlenecks](https://howaiworks.ai/blog/training-vs-inference-two-economies.md) — updated 2026-07-13
  Training is network-bound and paid once. Serving is memory-bound and paid forever. The crossover point, and why the 2024 chip-demand call was backwards.
- [Transformers.js v4: Revolutionizing Web-Based AI](https://howaiworks.ai/blog/transformers-js-v4-release.md) — updated 2026-04-02
  Hugging Face releases Transformers.js v4 with a new WebGPU runtime, drastic performance improvements, and support for massive models directly in the browser.
- [Transformers v5: PyTorch-First Library Update](https://howaiworks.ai/blog/transformers-v5-release-announcement-2025.md) — updated 2025-12-05
  Hugging Face releases Transformers v5 with PyTorch-only backend, quantization as first-class feature, and enhanced interoperability across the AI ecosystem.
- [Turbovec: Google's TurboQuant in an Open-Source Rust Index](https://howaiworks.ai/blog/turbovec-turboquant-vector-search-library.md) — updated 2026-08-25
  Turbovec is an MIT-licensed Rust vector index built on Google Research's TurboQuant quantizer, claiming 8x compression and faster search than FAISS.
- [UBTech Walker S2: $37M Border Patrol Robot Deal](https://howaiworks.ai/blog/ubtech-walker-s2-border-patrol-deployment-2025.md) — updated 2025-11-26
  UBTech Walker S2 humanoid robots are deployed for border patrol, demonstrating the feasibility of autonomous robotic systems in complex outdoor environments.
- [UBTech's Walker S2: $112M in Factory Orders](https://howaiworks.ai/blog/ubtech-walker-s2-factory-orders-2025.md) — updated 2025-11-14
  UBTech Robotics secures $112 million in orders for Walker S2 humanoid robots as Chinese factories adopt human-shaped automation for industrial tasks.
- [UltraData-Math: Scaling High-Quality Mathematical Reasoning](https://howaiworks.ai/blog/ultradata-math-openbmb-announcement.md) — updated 2026-02-19
  OpenBMB releases UltraData-Math, a 290B+ token dataset with a unique tiered grading system to boost LLM performance in complex mathematical tasks.
- [UMA Launches Physical AI Robotics from Europe](https://howaiworks.ai/blog/uma-launches-physical-ai-robotics-2025.md) — updated 2025-12-05
  UMA, founded by Tesla and Google DeepMind veterans, launches to build humanoid robots for real-world deployment in warehouses, hospitals, and factories.
- [URKL: World's First Humanoid Robot Combat League Opens in Shenzhen](https://howaiworks.ai/blog/urkl-humanoid-robot-combat-league-shenzhen-2026.md) — updated 2026-07-21
  EngineAI's Ultimate Robot Knock-out Legend puts 32 teams on identical T800 humanoids to fight for a $1.44M gold belt — a benchmark for embodied AI control.
- [VibeVoice-ASR: Long-Form Speech Breakthrough](https://howaiworks.ai/blog/vibevoice-asr-microsoft-speech-recognition.md) — updated 2026-01-26
  Discover VibeVoice-ASR, Microsoft's new model capable of 60-minute single-pass speech-to-text with speaker diarization and custom hotwords.
- [Vint Cerf Backs DNSid, a Plan to Identify AI Agents Online](https://howaiworks.ai/blog/vint-cerf-dnsid-agent-identity-2026.md) — updated 2026-07-21
  TCP/IP co-architect Vint Cerf is advising Innovation Labs on DNSid, a DNS-anchored standard, now at the IETF, that would give AI agents verifiable identities.
- [Waymo Brings Autonomous Rides to London in 2026](https://howaiworks.ai/blog/waymo-london-announcement-2025.md) — updated 2025-10-15
  Waymo announces its expansion to London, marking its first international deployment of autonomous ride-hailing services using the fifth-generation Waymo Driver.
- [Waymo World Model: Generative AI for Safer Autonomous Driving](https://howaiworks.ai/blog/waymo-world-model-announcement.md) — updated 2026-02-16
  Waymo introduces the Waymo World Model, a breakthrough generative AI built on Genie 3 that simulates hyper-realistic driving scenarios for safer navigation.
- [Welcome to HowAIWorks.ai](https://howaiworks.ai/blog/welcome-to-howaiworks.md) — updated 2025-08-01
  A reference site for people who use AI and want to understand it: a glossary, model and tool catalogs, practical use-case guides, and a blog.
- [What Is Quark? Alibaba's AI Super App, Explained](https://howaiworks.ai/blog/what-is-quark-ai-alibaba.md) — updated 2026-07-12
  Quark is Alibaba's Qwen-powered AI super app with ~159M monthly users. What it does, how to access it outside China, and how it compares to ChatGPT.
- [What One AI Query Costs: Tokens, Bytes, Joules, Cents](https://howaiworks.ai/blog/what-one-ai-query-costs.md) — updated 2026-07-13
  One question, one model, one machine — costed end to end. Then four dials that move the same answer by more than 10,000x.
- [White House Proposes 90-Day Pre-Release AI Model Testing](https://howaiworks.ai/blog/white-house-ai-model-testing-proposal.md) — updated 2026-05-22
  The US administration proposes a voluntary 90-day review period for flagship AI models, prompted by cybersecurity concerns and Anthropic's Mythos model.
- [Why NVIDIA's Moat Is Software, and What Would Break It](https://howaiworks.ai/blog/why-nvidias-moat-is-software.md) — updated 2026-07-13
  Rivals have shipped comparable silicon for years without shifting the market. The moat is the software above the chips and the rack around them.
- [World Labs Announces Atlas, an Omni World Model](https://howaiworks.ai/blog/world-labs-atlas-world-model-2026.md) — updated 2026-09-02
  World Labs unveiled Atlas on September 1, 2026: an omni world model over text, images, video and 3D that generates a minute of 1440p camera-controlled video.
- [Grok 4.6, Three Weeks On: The Score Moved, The Model Didn't](https://howaiworks.ai/blog/xai-grok-4-6-analysis-2026.md) — updated 2026-09-06
  Grok 4.6 scored 61 on Artificial Analysis Intelligence Index v4.1 and 51 on v4.2. Same weights. Plus the 200K billing cliff and the Cursor acquisition.
- [xAI Launches Grok Voice Think Fast 1.0](https://howaiworks.ai/blog/xai-grok-voice-think-fast-1-0.md) — updated 2026-04-24
  xAI announces Grok Voice Think Fast 1.0, a new flagship voice model built for complex workflows, real-time reasoning, and zero added latency.
- [Xiaomi MiMo-V2.5: The Next Generation of Open Agentic Models](https://howaiworks.ai/blog/xiaomi-mimo-v2-5-agentic-models.md) — updated 2026-04-23
  Xiaomi announces the MiMo-V2.5 series, featuring flagship agentic performance and native omnimodality to rival frontier models like Claude 4.6 and GPT-5.4.
- [Xiaomi MiMo-V2: Three New State-of-the-Art AI Models](https://howaiworks.ai/blog/xiaomi-mimo-v2-ai-models.md) — updated 2026-07-08
  Xiaomi releases MiMo-V2, a groundbreaking AI trio featuring a 1T-parameter Pro model, an omnimodal agent, and a next-gen TTS system.
- [Xiaomi-Robotics-0: Scaling VLA Models for Real-Time Robot Control](https://howaiworks.ai/blog/xiaomi-robotics-0-announcement.md) — updated 2026-02-16
  Xiaomi announces Robotics 0, a new division focused on building a universal robot operating system and affordable humanoid robots for home and industrial use.
- [Youtu-VL: Unified Vision-Language Supervision](https://howaiworks.ai/blog/youtu-vl-announcement.md) — updated 2026-07-21
  Tencent Youtu Lab introduces Youtu-VL, a 4B parameter model that pioneers the 'vision-as-target' paradigm for advanced visual perception.
- [Zoom AI Hits 48.1% on Humanity's Last Exam](https://howaiworks.ai/blog/zoom-humanitys-last-exam-breakthrough-2025.md) — updated 2025-12-15
  Zoom's federated AI approach achieves 48.1% on the rigorous Humanity's Last Exam benchmark, surpassing Google Gemini 3 Pro.