Mark Zuckerberg posted on X for the first time in more than three years on July 9, choosing Elon Musk’s platform over Meta’s own Threads to announce Muse Spark 1.1, a new agentic AI model, and the Meta Model API — Meta’s first paid platform allowing outside developers to build on its models.
Zuckerberg’s last post on X dated back to July 2023, making the choice of venue as much a signal as the announcement itself. His post read: “The Meta Model API allows developers to build using Muse Spark for the first time. Our focus is on delivering strong agentic and multimodal models at very low cost. More to come soon.” The original post is here.
Undercutting Anthropic and OpenAI on Price
Meta priced the Muse Spark 1.1 API aggressively at $1.25 per million input tokens and $4.25 per million output tokens — positioning it well below Anthropic’s listed rates for Claude Sonnet 5 and landing closer to the cost of Anthropic’s cheapest model, Claude Haiku 4.5. Meta had built AI models before, chiefly the open-weight Llama family, but had never previously charged developers directly for API access to a Meta model.
On capability, Meta claims Muse Spark 1.1 handles long-running tasks with a 1-million-token context window, supports parallel execution through sub-agents, and has been trained to operate computer interfaces across desktop, mobile and web browsers.
The company touted benchmark results including a score of 88.1 on MCP Atlas for scaled tool use, 54.7 on JobBench for professional tool use, 62.1 on Humanity’s Last Exam, and 57.2 on Finance Agent v2.
A New Front in the Model Wars
The launch puts Meta squarely into the paid developer-API business for the first time, competing directly with OpenAI, Anthropic, and Google for the attention of companies building AI agents and coding tools.
Coming just a day before Meta separately confirmed plans for “Meta Compute” — a broader cloud business reselling GPU capacity and hosted models — the API launch looks less like an isolated product drop and more like the opening piece of a wider commercial AI platform Meta is assembling around its Llama and Muse model families.
The timing is notable: Muse Spark 1.1 arrives in the same week as GPT-5.6 and Grok 4.5, making mid-July 2026 one of the most crowded stretches of frontier-model releases to date. Meta is betting that aggressive pricing, rather than topping benchmark leaderboards outright, will be its wedge into a developer market currently dominated by OpenAI and Anthropic.
Meta has not disclosed adoption figures for the new API since launch, and said only that “more” agentic and multimodal releases are coming soon.
Quick Links:
