Meta Opens Paid API With Muse Spark 1.1, Its Most Capable Agentic Model Yet
News Summary
Meta rolled out Muse Spark 1.1 on July 9, 2026 (Eastern Time), marking the company's first paid developer offering through its new Meta Model API and its most capable multimodal reasoning model to date. Built by Meta Superintelligence Labs under chief AI officer Alexandr Wang, the update arrives just three months after the original Muse Spark debuted in April 2026, and it is aimed squarely at agentic coding, planning, and computer-use tasks where developers ask an AI system to operate across multiple applications on its own.
What's New in This Release
Muse Spark 1.1 extends its predecessor with stronger performance on real-world software engineering tasks, including diagnosing and fixing bugs in large, complex codebases, implementing new features in enterprise-grade systems, and carrying out large-scale code migrations. Meta says the model can actively manage a context window of up to one million tokens, allowing it to keep track of long, evolving projects without losing earlier context. The company also highlighted improved performance on computer-use workflows, where the model navigates unfamiliar interfaces, adapts to changing information, and maintains state across extended, multi-step sessions.
Availability and Pricing
For the first time, Meta is charging outside developers for direct access to one of its models. Muse Spark 1.1 is available through a public preview of the Meta Model API, priced at $1.25 per million input tokens and $4.25 per million output tokens (Pacific Time launch window), which Meta describes as aggressively competitive with offerings from OpenAI and Anthropic. Consumers can already try the model for free in "Thinking" mode inside the Meta AI app and on meta.ai. Meta has not announced a date for general availability of the developer API beyond the current preview.
Where the Model Fits
Muse Spark already powers Meta AI across the company's consumer surfaces, including the Meta AI app and meta.ai, and is being extended to WhatsApp, Instagram, Facebook, Messenger, and Meta's AI-enabled glasses. Meta says the 1.1 update also improves recommendation and content features across Instagram, Facebook, and Threads. Industry benchmarking cited by several outlets suggests the updated model outperforms Google's latest Gemini release on coding and reasoning evaluations, underscoring how quickly the competitive gap between major AI labs continues to narrow.
Why It Matters
By opening a paid API tier, Meta is stepping directly into the commercial AI infrastructure market long dominated by OpenAI, Anthropic, and Google, giving developers another large-context, agent-capable model to build on. For the broader AI research community, the release signals that agentic capabilities such as long-horizon planning, tool use, and multi-app orchestration are becoming a central axis of competition among frontier labs, alongside raw reasoning and coding benchmarks.