Meta Muse Spark 1.2: A Coding-Focused AI Model Built for Real Developer Workflows

Meta has released Muse Spark 1.2, a significant coding-oriented update to its Muse Spark line, designed to handle complex software engineering tasks more effectively. Released in early August 2026 alongside the beta of Muse Code (a terminal-based coding agent), the model improves code generation, complex debugging, codebase understanding, and end-to-end developer workflows.

Meta significantly scaled training compute on coding tasks and expanded environment diversity. The model excels at long-horizon work—such as whole-repository generation, large multi-file projects, and sustained auto-research—using planning, goal conditioning, and context compaction to stay on track over extended sessions. It features a 1-million-token context window and multimodal support (text, images, video, PDFs, and more), enabling capabilities like turning a home fly-through video into a functional marketing and booking webpage.

In Meta’s evaluations, Muse Spark 1.2 reaches 82.9% on Terminal-Bench 2.1 and 59.3% on DeepSWE v1.1 (improvements over 1.1), while remaining competitive on internal coding benchmarks. It is available via the Meta Model API and powers Muse Code, which installs easily on macOS and Linux and supports persistent background agents, transparent event logging, and complex repository-scale tasks. Pricing includes a Standard tier ($1.25 per million input tokens / $4.25 output) that does not use your data for training, plus a much lower-cost Contributor tier.

Independent tests and practical demos (including code reviews, SDK readiness checks, and document generation) show it handling medium-to-complex coding projects efficiently and cost-effectively. For developers seeking strong coding performance, long-context agentic capabilities, and accessible pricing, Muse Spark 1.2 represents a compelling option in Meta’s evolving model lineup.