Blog
The pieces that do not fit in a glossary entry: where a benchmark number came from, what a claim actually shows, and what something costs in production.
Sonnet 5.5 vs Opus 5.5: Half the Price, Two Elo Points Behind
Opus 5.5 undercuts Opus 5 and Fable 5.1; Sonnet 5.5 keeps Sonnet 5's price and nearly ties it. Pricing, benchmarks and when the flagship is worth it.
Grok 4.6, Three Weeks On: The Score Moved, The Model Didn’t
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.
GPT-6 Astra Has No Realtime API and a 272K Pricing Cliff
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.
Nemotron 3.5 Lightning: One GPU, Yes. A Laptop, No.
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.
Gemini 3.8 Flash: Same Price Per Token, 40% More Per Task
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.
Claude Fable 5.1: Same $10/$50, Cache Reads Cut 75%
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.
The Newest Wan You Can Download Is Still Wan 2.2
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.
Qwen3.8-Max’s Open Weights Are Not the Model You Benchmarked
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.
Qwen’s E-Commerce Bench Runs AI Agents for a Simulated Year
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.
OpenAI Launches GPT-6 Astra With Record Benchmarks
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.