Meta prices Muse Spark 1.1 API at $1.25/$4.25 per M tokens

Meta finally attached a price tag to one of its models. The Meta AI blog announced Muse Spark 1.1 on July 9, 2026, framed as a multimodal reasoning model with a 1 million token context window, built for agentic tasks, tool use, computer use, and coding. What matters more than the model card is the second thing it shipped alongside: a public preview of the Meta Model API, which the post describes as OpenAI-compatible with structured output and parallel tool calling.

The pricing is the interesting part. According to CNBC, Meta will charge $1.25 per million input tokens and $4.25 per million output tokens, with $20 in free credits for every new API account, and the preview is limited to US developers at launch. Mark Zuckerberg described it as "the first time that we're doing a real serious API" and said "the pricing is going to be very aggressive and attractive." AI chief Alexandr Wang used the same phrase, positioning Muse Spark 1.1 against similar offerings from Anthropic and OpenAI, and called it Meta's "strongest model for agentic and coding work yet."

Why this matters if you are not shopping API prices: this is the first time Meta has charged businesses for access to its models, and it lands roughly a year after Zuckerberg brought Wang in to run Meta Superintelligence Labs. The company is picking price and OpenAI-compatibility as its wedge into the coding-agent market rather than raw benchmark leadership. The Meta blog leads with endorsements from Replit CEO Amjad Masad, Cline CEO Saoud Rizwan, and Box's Yashodha Bhavnani, which reads like a company that wants integrations, not headlines, as the early signal.

The honest caveat is that the blog does not publish specific benchmark scores against GPT or Claude. Wang's "strongest" framing and Masad calling it a "complete agentic foundation" are launch-day language from interested parties, and the coding-agent segment has a habit of eating those claims. What the reporting also does not tell you is the rate limits, how the 1M-token context is priced at length, or when developers outside the US get in.

If Muse Spark 1.1 holds up on independent coding evals, cost-sensitive teams already writing to OpenAI-shaped clients get something they have not really had from a hyperscaler yet, a genuine drop-in cheaper option. That is the pitch worth watching.