Meta Released Muse Spark 1.1 with Major Upgrades

Meta didn't bury the lede here. Muse Spark 1.1 landed through a brand-new public Meta Model API, and the specs read like a direct challenge rather than an incremental patch. A 1 million token context window, tool use, computer use, real coding gains, and multimodal reasoning all bundled into one release, aimed squarely at the coding-agent market that OpenAI and Anthropic have been treating as their own turf.

The pricing is the part that stings. Meta set Muse Spark 1.1 below what many frontier rivals charge, which is the kind of move that forces competitors to either match the discount or explain why they won't. It's a familiar playbook, but Meta hasn't really run it in this specific arena before, at least not with a model this capable attached to it.

Executives leaned into the framing. Alexandr Wang and Claire Zhou both pushed the release, and the coverage that followed, from CNBC, Bloomberg, and Axios, all landed on the same read: this is Meta's most direct shot yet at the coding-agent dominance OpenAI and Anthropic have enjoyed. Simon Willison tracked the release closely enough to be one of the go-to sources on it, Constellation Research weighed in with its own analysis, and the developer docs plus a busy Hacker News thread gave the technical crowd plenty to argue over.

What makes this notable isn't just the spec sheet. It's timing. Coding agents have become the battleground where labs prove their models can do real work, not just answer questions well. Meta showing up with a cheaper, long-context, tool-capable model changes the calculus for anyone building agent products on top of someone else's API. Whether developers actually switch is a separate question from whether they now have a reason to consider it, and Meta just gave them one.