📝 TL;DR 📝 Meta released Muse Spark 1.1 on July 9, its second model from Meta Superintelligence Labs, and for the first time ever opened a paid developer API for one of its own models. It is positioned as a frontier-tier agentic and coding model at roughly a quarter of the price of comparable Claude or GPT models. It is free to try in the Meta AI app today. Independent testing suggests it genuinely leads on agentic tool-use tasks while trailing on pure coding, and Meta's own benchmark numbers do not always match what independent evaluators are finding. 🧠 Overview 🧠 This landed in an unusually crowded 24-hour stretch for AI releases, alongside OpenAI's GPT-5.6 rollout and SpaceXAI's Grok 4.5 launch. Of the three, Muse Spark 1.1 is arguably the most strategically significant for Meta specifically, because it represents a real reversal of the company's AI strategy. Until this week, Meta's entire public AI identity was built around giving models away for free as open weights through the Llama family, which has been downloaded more than 1.2 billion times. Muse Spark 1.1 is closed, hosted, and metered per token, putting Meta directly into the same paid-API business Anthropic and OpenAI have run for years. Mark Zuckerberg personally announced the launch on X, ending a three-year absence from the platform to do it, a detail several outlets read as a signal of how much internal weight Meta is putting behind this release. 📜 The Announcement 📜 Muse Spark 1.1 is a multimodal reasoning model built specifically for agentic tasks: tool use, computer use, coding, and multi-agent orchestration, with a self-managed one-million-token context window, up from roughly 260,000 tokens in the original Muse Spark from April. It is free to use in "Thinking" mode inside the Meta AI app and at meta.ai with a Meta login. For developers, the new Meta Model API launched in public preview on July 9, priced at $1.25 per million input tokens and $4.25 per million output tokens, alongside $20 in free credits for new accounts. The API is OpenAI and Anthropic SDK compatible, meaning developers can typically point an existing agent at Muse Spark with a base URL and key change rather than a full rewrite.