MENLO PARK, Calif.: Meta releases Muse Spark 1.1 and, for the first time, charges developers to use one of its models.
The new Meta Model API opens in US-only public preview. Pricing runs $1.25 per million input tokens and $4.25 for output. New accounts get $20 in free credits.
This is a real break from Meta’s history. The company built its AI reputation on giving Llama away. Charging for API access says Meta now thinks Muse Spark can compete at the frontier on its own terms.
The model is built for agent work. It carries a self-managed one-million-token context window, runs computers across desktop, browser, and mobile, and delegates to parallel subagents. MCP and custom skills come standard.
Meta’s own benchmarks put it first on MCP Atlas at 88.1, plus JobBench, Humanity’s Last Exam, and Finance Agent V2. On the held-back Vals AI Harvey legal benchmark it scores 20% against Fable 5’s 11%. It still trails Claude Opus 4.8 and GPT-5.5 on straight SWE-Bench coding. Independent verification hasn’t landed yet, so treat the chart with appropriate salt.
One clever touch: the API speaks both the OpenAI and Anthropic SDK formats. Pointing an existing agent at Muse Spark is a base-URL and key change. Replit, Cline, and Box are early partners.
Consumers get the model free through Thinking mode in the Meta AI app. Broader deployment across WhatsApp, Instagram, Facebook, and smart glasses follows.
Usage-based billing starts once the introductory credits run out. The launch week was absurd even by 2026 standards. OpenAI’s GPT-5.6 family and xAI’s Grok 4.5 shipped within the same 48 hours.
New Delhi: Adobe has acquired Rilo, an India based marketing intelligence startup, in a deal…
New York: Anthropic has launched Claude Fable 5.1 and Claude Mythos 5.1, its latest AI…
Bengaluru: Perplexity AI has launched a new product called Portable Computer, an AI agent that…
New Delhi: Bengaluru-based voice AI startup Ringg has raised $10 million from Peak XV Partners,…
New York: Google is moving forward with Android Halo, a new system-level feature designed to…
New York: Users across Anthropic’s Claude platform are experiencing elevated errors affecting multiple Claude models,…