Meta opens Muse Spark AI model API to developers — and it's already beating Llama

Meta quietly turned on the spigot Thursday. After months of private testing with a handpicked set of partners, the company’s Muse Spark AI model is now available to any US developer through the Meta Model API public preview — complete with an upgraded 1.1 version that the company says is its strongest model yet for real-world coding and AI agent tasks.

Muse Spark first appeared in April, the debut model from Meta’s “Superintelligence” team formed last year. At launch it was invite-only, a cautious rollout that let Meta test the waters while it scrambled to close the gap with OpenAI and Anthropic. Thursday’s public release — alongside Muse Spark 1.1 — is the clearest signal yet that Meta believes the model is ready for prime time.

The 1.1 version packs improvements across four areas: writing and debugging code, operating software and calling external tools, understanding text-image-video inputs, and completing complex multi-step tasks with less human hand-holding. Meta frames it as a pillar of its “Personal Superintelligence” strategy — a phrase that sounds grandiose until you look at what the model is actually doing inside Meta’s own products.

Muse Spark 1.1 is already live in Thinking mode inside Meta AI’s app and web interfaces. And the company plans to eventually phase out the Llama models that currently power WhatsApp, Instagram, Facebook, and Meta’s smart glasses chatbot, replacing them with Muse Spark. That’s a non-trivial migration — Llama is the backbone of Meta’s consumer AI presence, and swapping it out signals real confidence in the new model.

Pricing lands somewhere in the middle of the market. Meta charges $1.25 per million input tokens and $4.25 per million output tokens — pricier than OpenAI’s entry-level GPT-5 mini and Anthropic’s budget-friendly Claude Haiku 4.5, but cheaper than Anthropic’s high-end Claude Sonnet 4.6. New developers get $20 in free credits to kick the tires.

The API rollout follows a busy week for Meta’s AI division. On Tuesday the company expanded generative AI features across its apps and introduced Muse Image, the first image-generation model from Meta Superintelligence Labs. Between the two launches, Meta is making clear it intends to be more than just an open-source AI player — it wants to compete toe-to-toe in the paid model market too.