Meta Launches Muse Spark 1.1 Paid API; Charges $1.25/$4.25 Per Million Tokens
Meta released Muse Spark 1.1 on Thursday with public developer API access, marking the company's first paid AI model tier and direct competition with Anthropic and OpenAI in proprietary model sales. The updated model is positioned as Meta's strongest for agentic and coding tasks, with support for 1 million-token context windows and native multimodal reasoning across text, images, and video. Pricing is $1.25 per million input tokens and $4.25 per million output tokens, cheaper than OpenAI's Grok 4.5 and Anthropic's Opus. New API accounts get $20 in free credits to start.
Muse Spark 1.1 supports tool use and computer automation for orchestrating multi-step workflows across applications. The company trained the model to work with OpenClaw and other popular agentic harnesses to maximize adoption in the emerging agent-loop ecosystem. Meta is limiting API access to its own infrastructure initially rather than third-party marketplaces like OpenRouter, giving the company direct visibility into usage patterns. A lower-cost open-source variant is in development but no release date was given. Meta is also training Watermelon, a more powerful next-generation model slated for later in 2026.
For teams evaluating agent platforms, Meta's aggressive pricing on Muse Spark 1.1 makes it worth benchmarking against Anthropic and OpenAI for routine coding and agentic workloads where the model can deliver GPT-5.5-class capability at lower token cost. The 1M token context window and native multimodal support fit the rising demand for agents that reason over codebases, documents, and visual context. Meta is betting that controlling the full infrastructure stack—compute, model, and API—lets it undercut pure model-provider pricing while capturing margin on infrastructure.