Meta unveils Muse Image generator and Muse Spark 1.1 coding model

Meta's Superintelligence Labs introduced two new AI models. The Muse Spark 1.1 language model scored near the top of the Artificial Analysis Intelligence Index, reaching 51 points overall and 71.3 points in the coding sub‑test, out‑performing GLM‑5.2 while costing roughly $0.26 per test task—significantly cheaper than rivals. It also reduced hallucinations to 38 % by opting to remain silent when uncertain and expanded its context window to one million tokens. The model is currently accessible only through Meta’s own API, limiting broad adoption.

In parallel, Meta announced Muse Image, its first image‑generation and editing model from the same lab. Muse Image can create new pictures and modify existing ones via natural‑language prompts, supporting multi‑step edits, style changes, and text insertion. The system is integrated with Meta’s platforms, including Instagram Stories and WhatsApp chats, and offers preset templates for common tasks such as photo restoration and stylized portraits. Global rollout is pending, with availability initially limited to selected markets.