Meta Launches Muse Spark 1.1 With 1 Million Token Context

Meta has announced Muse Spark 1.1, a major upgrade to the original Muse Spark model released in April. The new model was developed by Meta Superintelligence Labs, the group Meta created to work on advanced AI systems.

The company says Muse Spark 1.1 improves several key areas, including coding, multimodal reasoning, long-context work, and tool-based agentic tasks.

Muse Spark 1.1 focuses on agentic workflows

Muse Spark 1.1 can use native tools, MCP servers, and custom skills. Meta says the model can act as a primary agent, create a plan, and assign tasks to multiple subagents.

Those subagents can work in parallel, which should help the model complete more complex workflows faster. Meta is also positioning Muse Spark 1.1 as a stronger model for computer-use tasks, where AI systems interact with apps and interfaces to complete user requests.

Meta adds a 1 million token context window

One of the biggest upgrades is the new 1-million-token context window. This gives Muse Spark 1.1 far more room to process large documents, long conversations, codebases, and complex project instructions.

Meta says the model does more than store large amounts of information. According to the company, Muse Spark 1.1 can actively manage its context during long workflows, which should make it more useful for extended tasks that require memory, planning, and follow-through.

Muse Spark 1.1 targets frontier benchmark performance

Meta says Muse Spark 1.1 reached state-of-the-art results on MCP Atlas, JobBench, Humanity’s Last Exam, and FinanceBench.

The company also claims the model competes with more expensive frontier models, including GPT-5.5 and Claude Opus 4.8, on other benchmarks.

These results suggest Meta wants Muse Spark 1.1 to compete directly with leading AI models used for coding, reasoning, enterprise workflows, and research-heavy tasks.

Meta Model API opens to developers

Muse Spark 1.1 is also Meta’s first model available through a public API. Developers can access it through the new Meta Model API.

At launch, new Meta Model API accounts receive $20 in free credits. Pricing starts at $1.25 per million input tokens, while output tokens cost $4.25 per million.

Meta says this makes Muse Spark 1.1 substantially cheaper than competing frontier models from OpenAI and Anthropic.

Meta says Muse Spark 1.1 passed safety checks

Meta evaluated Muse Spark 1.1 under its Advanced AI Scaling Framework. The assessment covered frontier risk areas, including cybersecurity, chemical and biological risks, and loss of control.

The company says the model stayed within its defined safety margins across those categories.

Muse Spark 1.1 is available now

Muse Spark 1.1 is available now in Thinking mode. Consumers can access it through the Meta AI app and the meta.ai website. Meta is also reportedly working on another model codenamed Watermelon.

The launch comes as competition in AI models continues to intensify. OpenAI has also launched GPT-5.6 to the public and introduced ChatGPT Work, giving users more options for advanced reasoning, coding, and agentic workflows.