Meta released Muse Spark 1.1 on July 9, 2026, and for the first time opened one of its in-house foundation models to outside developers through a new Meta Model API. The API is in public preview for US developers starting today, with $20 in free credits attached to every new account. The move ends a stretch in which Muse Spark was accessible only inside Meta's own surfaces, and puts the company into direct commercial competition with the API businesses run by OpenAI, Google, and Anthropic.
The 1.1 upgrade is pitched primarily at coding and agent developers. Meta says Muse Spark 1.1 can detect and fix complex bugs, support end-to-end agentic workflows including multi-agent systems, and handle native multimodal perception across images, video, and documents. Meta describes the new model as a "step-change" from the April release, with improvements shaped by developer feedback gathered since the first Muse Spark shipped.
The timing matters. Muse Spark first launched in April 2026 as Meta's belated re-entry into the frontier-model race, then progressively took over the chatbots inside the Meta AI app, Instagram, WhatsApp, and the latest generation of Meta smart glasses. Opening a public API is the step that turns Muse from an internal product feature into a platform bet — and forces it to be measured against the incumbents on price, latency, and coding benchmarks rather than on default distribution inside Meta's apps.
Key facts
- 01Meta released Muse Spark 1.1 on July 9, 2026, its first Muse model available through a public developer API.
- 02The Meta Model API is in public preview for US developers, with $20 in free credits per new account.
- 03Muse Spark 1.1 adds agentic coding, multi-agent workflow support, and native multimodal perception across images, video, and documents.
- 04The launch follows Muse Image, released earlier in the week, which drew criticism for incorporating other users' Instagram content into generations.
- 05The first Muse Spark model launched in April 2026 and now powers chatbots in the Meta AI app, Instagram, WhatsApp, and Meta smart glasses.
Muse Spark 1.1 is available immediately in Thinking mode through the Meta AI app and the Meta AI website, in addition to the API preview. Thinking mode is Meta's branded reasoning setting, designed for problems where longer deliberation produces better answers — the same category of feature OpenAI, Google, and Anthropic have all built out over the past year. Meta has not disclosed pricing beyond the $20 credit grant, but the credit is a familiar mechanism for pulling developers off competing APIs long enough to benchmark.
The launch lands 48 hours after Muse Image, Meta's new image-generation model, went live and immediately drew criticism for a feature that lets outputs incorporate other users' Instagram content into generations. Muse Image and Muse Spark are separate models but share the Muse brand and the same underlying justification: Meta needs shipped products to defend the tens of billions of dollars it has poured into AI infrastructure, chip supply, and a hiring spree that reshaped the company's org chart last year.
Coding is a deliberate choice of battleground. It is the AI segment where developer revenue is easiest to attribute, where benchmarks are public and adversarial, and where switching costs between model providers are low — a single API endpoint change is often the entire migration. Meta is not the first entrant here; Anthropic's Claude models dominate paid coding workloads, and OpenAI's GPT line still holds the largest developer base. Meta has to prove Muse Spark 1.1 is competitive on real repositories, not just on cherry-picked demos.
The agentic pitch is where Meta is trying to leapfrog. Multi-agent orchestration — one model coordinating others, calling tools, and holding state across long-running tasks — is the workload most enterprise customers are actively piloting in 2026, and it is where reliability, not raw intelligence, decides who wins the contract. Meta claims Muse Spark 1.1 supports end-to-end agentic workflows across a range of apps, but the company has not published benchmark numbers or a system card alongside the launch.
That is the caveat worth flagging. Meta has released capability claims without the supporting evals that OpenAI, Anthropic, and Google now routinely publish for their frontier releases — no SWE-bench score, no agent-benchmark result, no third-party red-team. Developer trial usage will fill that gap quickly once the API is live, but until independent numbers land, the "step-change" framing is a marketing claim rather than a measured one. The Instagram-content controversy attached to Muse Image also puts every subsequent Muse launch under closer scrutiny on training-data provenance.
Meta's shift from closed model to public API is the substantive story here. Owning distribution through Instagram, WhatsApp, and the Meta AI app is a powerful moat for consumer AI, but it does not generate the developer-driven revenue that has made OpenAI and Anthropic the fastest-scaling software companies in history. By pricing Muse Spark 1.1 into the API market and targeting coding first, Meta is signaling that model-as-a-service is now a business line, not a research demo — and the next signal to watch is what the per-token pricing looks like when the free credits run out.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
