# Meta Starts Charging for Its Own Model With Muse Spark 1.1 and a New API

The agent-focused model comes with a one-million-token context window and OpenAI- and Anthropic-compatible endpoints, priced at $1.25 and $4.25 per million input and output tokens.

- Published: 2026-07-12T05:29:34.912Z
- Canonical: https://polylog.news/ai/2026-07-12/meta-starts-charging-for-its-own-model-with-muse-spark-1-1-a
- Publisher: Polylog (AI desk)
- Section: tech
- Sources: [Meta AI](https://ai.meta.com/blog/introducing-muse-spark-meta-model-api/)

Meta released Muse Spark 1.1 on July 9 and opened a public preview of the Meta Model application programming interface (API), the first time the company has charged developers to use its own model rather than releasing open weights. The [announcement](https://ai.meta.com/blog/introducing-muse-spark-meta-model-api/) positions the model for agentic and coding work, with a one-million-token context window, multimodal input across images, video, and documents, built-in search with citations, structured output, and parallel tool calling.

The commercial framing matters as much as the specifications. The API supports the OpenAI Chat Completions and Responses formats and the Anthropic Messages format, so redirecting an existing agent is a base-URL and key change rather than a rewrite, per Meta's [developer materials](https://developer.meta.com/ai/resources/blog/build-with-muse-spark/). Pricing is $1.25 per million input tokens and $4.25 per million output tokens, with $20 in starter credits, placing it below the top closed frontier tiers, as [TechCrunch](https://techcrunch.com/2026/07/09/meta-enters-the-crowded-ai-coding-battle-with-muse-spark-1-1/) noted.

For a company that built its AI standing on downloadable Llama weights, a paid, closed API is a strategic pivot toward the coding and agent market that Anthropic and OpenAI already contest. Meta has not published independent third-party coding benchmarks alongside the release, so the "top-tier coding" claim rests on vendor framing until outside evaluations appear.

## What this means

Meta is trading some of the open-weight distribution that made Llama widespread for direct API revenue and control, betting the coding and agent segment is worth monetizing. The exposure is competitive. By replicating OpenAI and Anthropic wire formats and undercutting on price, Meta targets developers already dependent on those software development kits (SDKs), pressuring incumbents on cost per token rather than raw capability.

## What to watch

- Independent coding benchmarks such as SWE-bench Verified run by outside labs, which will show whether Muse Spark 1.1 actually competes with Claude and GPT models or only on price.
- Whether Meta keeps releasing open weights for its base models alongside the paid API, which will signal how far it is reversing its open-source positioning.