Meta released Muse Code (public beta) and Muse Spark 1.2 on August 5, 2026, its entry into the dedicated AI coding-agent market. Muse Code is a terminal agent that plans changes, writes code, and validates results across large repositories on macOS and Linux, powered by Muse Spark 1.2, a coding-focused model upgrade to Muse Spark 1.1. The model was trained heavily on long-running programming tasks including full repository generation and independent research, with end-to-end developer workflows in mind.
Standard API pricing for Muse Spark 1.2 is $1.25 per million input tokens and $4.25 per million output tokens—matching Muse Spark 1.1. A Contributor tier offers 10x cheaper access at $0.10 input and $0.20 output per million tokens, but requires users to share prompts and completions for Meta model training. Artificial Analysis estimates Muse Spark 1.2 at ~$0.69 per coding task on the Vals Index, making it 5th most capable coding model after Claude Opus 5, GPT-5.6 Sol, Kimi K3, and Claude Fable 5, at a cost roughly 3–20x lower than those alternatives depending on tier.
On Meta's own benchmarks, Muse Code achieved 82.9% on Terminal-Bench 2.1 (Meta's vendor-run measurement), though independent Vals testing rates it 14th under a common harness. The model trails Claude Code at 86.7% on the same benchmark family. Meta emphasizes agentic capabilities: Muse Spark 1.2 ranks #5 on GDPval-AA v2 (realistic knowledge-work orchestration), and Muse Code maintains four concurrent agent types—Coordinator, Explorer, Executor, Verifier—with persistent context across sessions, crash replay via event logging, and 1M token context window out of the box.