On July 9, 2026, Meta opened its Muse Spark 1.1 model to US developers in a public preview of the new Meta Model API. The company positions it as a tool that can plug into AI coding software and compete with OpenAI, Google, and Anthropic. It is the first time a Meta in-house model has reached external developers through an API.
Key takeaways
- Muse Spark 1.1 is available from July 9, 2026 through the new Meta Model API in public preview for US developers.
- Meta includes $20 of free credits with every new Meta Model API account.
- The model also runs in Thinking mode via the Meta AI app and the Meta AI website.
- New capabilities: detection and fixing of complex bugs, support for multi-agent workflows, and native multimodal perception across images, videos, and documents.
- The launch follows the Muse Image model, which drew controversy for pulling other Instagram users’ content into its generations.
What the "step-change" claim covers
Meta calls Muse Spark 1.1 a "step-change" over the first generation from April, built on developer feedback. The stated improvements cluster in three areas:
- More advanced coding, including detecting and fixing complex bugs.
- Better handling of end-to-end agentic workflows across many apps — including multi-agent systems.
- Native multimodal perception spanning images, videos, and documents.
It is worth separating claim from evidence. Meta did not publish specific benchmark results for the new capabilities at launch — the description stays at the level of announced abilities rather than confirmed, comparable numbers. Until independent tests confirm bug-detection rates or agent success, "step-change" is a vendor claim.
From an Instagram chatbot to a developer API
Muse Spark's trajectory matters. Meta released the first model in April 2026, reentering the AI race after a string of high-profile hires and a restructuring. It was initially available only directly through Meta AI, then began powering chatbots inside Instagram and WhatsApp and the latest Meta smart glasses. Version 1.1 is the first time the model moves beyond Meta's own apps and becomes a component an outside developer can wire into their tool.
The competitive context is unforgiving. The API opens the same week as the Muse Image model launch — and just as Anthropic ships Claude Cowork on mobile and web and OpenAI pushes its own agent. Meta is trying to justify the billions spent closing the gap and to reach parity with the leaders. A coding model in an API is a logical move: it is a segment where developers actually pay for tokens and where quality is easy to compare. The free $20 in credits is a classic trial incentive — a low barrier to pull in programmers accustomed to rivals.
Why it matters
Opening Muse Spark via an API is a shift in Meta's distribution model, not just another version of a model. Until now Meta kept its model inside its own apps, where it controlled the entire experience. Letting it into external coding tools means the company accepts competing in the open developer market, where quality and price win rather than social-platform reach. That is harder ground — developers compare models directly and drop the weaker one fast. The US-only regional limit and the absence of hard benchmarks suggest this is a cautious market test rather than a full rollout. If 1.1 genuinely matches the leaders on coding, Meta gains a new channel for revenue and usage data. If not, it confirms that capital alone does not buy quality parity in the most demanding segment.
What's next?
- The Meta Model API public preview is limited to US developers — a geographic expansion will signal production readiness.
- No published benchmarks for the new capabilities — independent tests of coding and agentic workflows will verify the "step-change" claim.
- Meta is building infrastructure for the Muse Spark series (7 GW of compute in 2026) — the compute scale will indicate how aggressively it intends to scale the model.
