Introduction
Meta Superintelligence Labs released Muse Spark 1.1 on July 9, 2026, a multimodal reasoning model built for agentic work, and opened it to developers through a new public API. This recap covers what shipped and why it matters. The release marks Meta's first paid AI model — the first time the company has charged businesses for API access to one of its models — priced at $1.25 per million input tokens and $4.25 per million output tokens.
Muse Spark 1.1 succeeds the original Muse Spark, internally codenamed Avocado, which shipped in April 2026 to a closed group of partners with no public API. Three months later, Meta is opening the model to any developer through a waitlisted portal, positioning it directly against OpenAI's and Anthropic's coding and agentic models on both price and capability.
What's New in Muse Spark 1.1
Meta describes Muse Spark 1.1 as a significant upgrade over the original release, with gains concentrated in four areas:
- Agent orchestration — the model can coordinate multi-agent systems, dispatching parallel subagents to optimize end-to-end latency on complex, multistep tasks.
- Computer use — it navigates across multiple applications and services, maintaining context over extended sessions and adapting as requirements change mid-task.
- Coding — performance improved substantially on real-world tasks involving large, complex codebases: diagnosing and fixing bugs, implementing features in enterprise-grade systems, and executing large-scale code migrations.
- Multimodal understanding — the model processes images, video, and audio, generates code from visual input, and produces detailed captions.
The model also generalizes zero-shot to new native tools, Model Context Protocol (MCP) servers, and custom skills, and supports structured output and parallel tool calling — capabilities aimed squarely at developers building AI agents rather than single-turn chat applications.
Pricing and API Access
The Meta Model API is now in public preview for US developers, distributed as an OpenAI-compatible package so existing tooling can point at it with minimal changes, and new accounts get $20 in free credits. Meta priced Muse Spark 1.1 at $1.25 per million input tokens and $4.25 per million output tokens. Zuckerberg framed that as roughly a quarter of what OpenAI and Anthropic charge for their top-tier flagship models; on a token-for-token basis, TechCrunch put it in line with — though slightly above — Claude Haiku 4.5 and OpenAI's limited-preview GPT-5.6 Luna.
That pricing is the more consequential part of the announcement: it's the first time Meta has asked businesses to pay for direct access to its models, giving the company a commercial AI revenue line separate from advertising and the free Meta AI consumer app. Muse Spark 1.1 also powers Meta AI in a new "Thinking" mode, available now in the Meta AI app and on meta.ai.
Context Window and Tool Use
Muse Spark 1.1 ships with a 1 million token context window, with what Meta calls active context management: rather than simply truncating old turns, the model compacts prior steps while retaining the information needed for later work in a long agentic session. Combined with the multi-agent orchestration features, this is aimed at tasks that run for many steps and across many tools — the kind of long-horizon coding and computer-use work that has become the standard benchmark for agentic models in 2026.
Early Reactions
Partners quoted in Meta's announcement were positive about the model's fit for production agentic systems. Replit's CEO called it "a complete agentic foundation," pointing to its "massive million-token context, full multimodal support, structured output, and parallel tool calling." Box said Muse Spark 1.1 delivers "enterprise capabilities competitive with today's leading frontier models."
Mark Zuckerberg posted about the release on X — his first post on the platform in three years — describing Spark as "a strong agentic and coding model at a very low price," and highlighting its agentic performance, tool use, and computer use. He also signaled that further model releases are coming from Meta Superintelligence Labs, the unit Alexandr Wang has led as Meta's chief AI officer since it was formed in mid-2025.
Meta says the model was evaluated under its Advanced AI Scaling Framework and found to operate within safe margins across chemical and biological, cybersecurity, and loss-of-control risk categories.
Conclusion
Muse Spark 1.1 is a modest-sounding point release with an outsized business implication: it's the model Meta chose to start charging for. By pairing a 1 million token context window and strong agentic and coding benchmarks with pricing well below OpenAI and Anthropic's top tiers, Meta is making an aggressive play for developers building agents and coding tools, not just for the consumer chatbot market it has focused on to date. The launch lands in a crowded month for agentic and coding model releases, and it's a signal that Meta intends to compete on API business, not just app usage.
Sources
- Introducing Muse Spark 1.1 (Meta AI blog)
- Meta enters the crowded AI coding battle with Muse Spark 1.1 (TechCrunch)
- Meta releases latest update of AI model Muse Spark (Fortune)
- Meta updates its Spark model, releases developer version (Axios)
- Meta Starts Charging for AI With Muse Spark 1.1 Agentic Model (Bloomberg)
- Meta Superintelligence Labs Releases Muse Spark 1.1 (MarkTechPost)