Meta Charges for Muse Spark 1.1 and Undercuts the Field

Meta released Muse Spark 1.1 on Thursday, and the story is not really the model. It is the price tag now attached to it. For the first time, developers who want to build on Meta's flagship model have to pay for the privilege, a decision that quietly closes the door on years of the company presenting itself as the open-weights champion of the industry.

The model itself is capable. Meta describes it as proficient at coding, video captioning and reasoning, and it ships with a self-managed one-million-token context window plus native support for agent and subagent orchestration, the Model Context Protocol and custom skills. Alexandr Wang, Meta's first chief AI officer and the head of its reorganised Meta Superintelligence Labs, wrote on X that Muse Spark 1.1 rivals GPT-5.5 and Opus 4.8 across a range of agentic evaluations.

The price is the message

Meta is charging $1.25 per million input tokens and $4.25 per million output tokens through its new Meta Model API, currently in public preview for developers in the United States. That sits below the comparable frontier tiers from OpenAI and Anthropic, and Mark Zuckerberg, who returned to X after three years to promote the launch, framed the pricing as deliberately aggressive.

The strategy is easy to read. Meta spent $14.3 billion in 2025 to acquire a 49 percent stake in Wang's data-labelling company, Scale AI, and folded his team into a new superintelligence unit. Having invested at that scale, the company wants revenue and reach now, not the goodwill that open weights once bought it. Underpricing the market is how a late entrant buys its way into developers' workflows.

Still chasing the leaders

Meta's own blog post claims Muse Spark 1.1 surpasses recent models from OpenAI, Anthropic and Google, and points to stronger coding and reasoning results against Google's latest Gemini. The independent picture is more measured. On at least one coding benchmark the model still trails Anthropic's Mythos 5 and Fable 5 as well as OpenAI's GPT-5.6, which suggests Meta is buying position with price while it works to close the gap on quality.

That places Muse Spark in a crowded week. It lands alongside the public launch of OpenAI's GPT-5.6 family and a fresh round of releases from nearly every major lab. It also marks a retreat from the open-source posture that defined Meta's earlier Superintelligence Labs work, and it arrives just as American companies are already turning to cheaper Chinese open models for the exact cost reasons Meta is now trying to answer.

Whether the bet works depends on developers who have spent two years treating Meta's models as something to download rather than rent. Muse Spark 1.1 asks them to change that habit. The pricing is the whole argument.