Meta opens Model API with Muse Spark 1.1
Meta released Muse Spark 1.1 on July 9 with a 1M-token context at $1.25/$4.25 per million tokens, and opened the Meta Model API to US developers.
Meta released Muse Spark 1.1 on July 9 and opened the Meta Model API to US developers in public preview the same day. The model carries a 1 million token context window, handles images, video, PDFs, and text, and is priced at $1.25/M input and $4.25/M output. New accounts get $20 in free credits.
Muse Spark 1.1 is the product of Meta Superintelligence Labs. It targets agentic tasks: multi-application computer workflows, tool use, coding, and multi-agent orchestration where the model delegates to subagents. Meta says the model can retain context across long sessions without manual summarization. Replit’s CEO called it “a complete agentic foundation.”
The Meta Model API uses an OpenAI-compatible interface. The model is also available in the Meta AI app and meta.ai in Thinking mode.
The previous Meta Muse Image release (covered July 8) was from the same Superintelligence Labs group but is a separate image-generation model. Muse Spark 1.1 is the text-and-multimodal reasoning counterpart.
What it means for you
Meta now has a direct API for developers. That did not exist before July 9. If you have been watching Llama weights without a hosted option, or have been evaluating whether Meta’s agentic capabilities were worth the infrastructure overhead, the public preview removes that barrier.
At $1.25/$4.25 per million tokens, Muse Spark 1.1 is priced below GPT-5.6 Terra ($2.50/$15) and below Grok 4.5 ($2/$6) on the input side. The 1M-token context is on par with Claude Sonnet 5 and above Grok 4.5’s current 500K. For agentic coding workloads, the benchmark picture is not yet clear; third-party evaluations are just beginning.