Meta Superintelligence Labs releases Muse Spark 1.1 with a 1M context window and a public Model API preview at $1.25/$4.25 — Meta's first move into the paid frontier tier alongside OpenAI and Anthropic.
On July 9, Meta Superintelligence Labs released Muse Spark 1.1 — a multimodal reasoning model built for agentic tasks with major gains in tool use, computer use, coding, and multimodal understanding over the April Muse Spark release. The model ships with a 1,048,576-token context window and runs in Thinking mode inside the Meta AI app and at meta.ai.
The structural shift is the Meta Model API public preview: developers access Muse Spark 1.1 at $1.25 per million input tokens and $4.25 per million output tokens, with cached input at $0.15 and web-search grounding at $2.50 per 1,000 queries. The API is drop-in compatible with OpenAI SDK, Anthropic SDK, and agent CLIs — base URL https://api.meta.ai/v1, model ID muse-spark-1.1. New accounts receive a one-time $20 free credit.
CEO Mark Zuckerberg posted on X for the first time in three years to promote the launch, calling Spark strongest at agentic performance, tool use, and computer use at a very low price. Reuters noted the pricing sits slightly above Claude Haiku 4.5 and GPT-5.6 Luna — a deliberate volume-market positioning as Google delays Gemini 3.5 Pro.
Our Take
Meta is no longer just an open-weights benefactor — it is now a paid API competitor. For enterprises under CFO scrutiny, Muse Spark 1.1's pricing makes agentic workloads economically viable at scale, but the closed-weights model means less auditability than Llama. Model-routing strategies should add Meta as a third production tier alongside OpenAI and Anthropic, with failover tested before budget cycles close.