Meta's Muse Spark 1.1 beats GPT-5.5 on agent evals, launches paid API
Zuckerberg drops Muse Spark 1.1: 1M-token agent model rivals GPT-5.5
Meta AI today unveiled Muse Spark 1.1, a major upgrade to its agentic model line, alongside the company's first-ever paid developer API. Announced by Mark Zuckerberg on X, Muse Spark 1.1 features a 1-million-token context window and achieves state-of-the-art results on agentic benchmarks, including MCP Atlas, JobBench, Humanity's Last Exam, and Finance Agent V2 — claiming the #1 spot ahead of GPT-5.5 and Claude Opus 4.8. Notably, on the held-back Vals AI Harvey legal-agent benchmark, it scores 20% against Fable's 11%, demonstrating strong domain-specific reasoning.
The accompanying Meta Model API is now in public preview, offering $20 in free credits for US developers. Pricing is set at $1.25 per million input tokens and $4.25 per million output tokens. The API supports computer use across desktop, browser, and mobile, as well as parallel subagent delegation — enabling complex multi-step automation. Early partners include Replit, Cline, and Box. Meta confirmed no open-weight release for this model, signaling a shift toward a more commercial, API-driven strategy for its frontier models.
- Muse Spark 1.1 features a 1M-token context window and rivals GPT-5.5/Opus 4.8 on agentic benchmarks.
- Meta's first paid developer API launches at $1.25/$4.25 per 1M tokens (in/out), with $20 free credits.
- Early partners: Replit, Cline, Box; supports computer use and parallel subagent delegation.
Why It Matters
Meta's paid API and top-tier agent model signal its serious push into commercial AI, challenging OpenAI and Anthropic.