Crypto
Home›Crypto›Market Structure›Meta launches Muse Code, a terminal coding agent in be…
Meta launches Muse Code, a terminal coding agent in beta
Meta says Muse Code uses a local replay-exact event log, so long jobs can restart after a crash, and it trails Claude Code on key benchmarks while beating others on most tests.
Meta has released Muse Code (beta), a terminal coding agent powered by its updated Muse Spark 1.2 coding model, with access via the Meta Model API and a curl install script, according to Decrypt.
The agent coordinates persistent background subagents and records a replay-exact event log, logging every model call, tool run, approval, and edit to a local file so that if the agent crashes it can resume precisely where it stopped. Meta said Muse Code is designed to take on complex software engineering tasks across large repositories, including planning changes, writing code, and validating results, with default commands like “/plan” for approval-gated planning, “/grill” to stress-test plans, and “/goal” to work toward completing objectives.
On performance, Decrypt reports that Meta’s own charts show Muse Spark 1.2 with Muse Code trailing Anthropic’s Claude Code on every coding benchmark shown, while it beats OpenAI’s Codex and Google’s Antigravity on most. On Terminal-Bench 2.1, Muse Spark 1.2 with Muse Code scored 82.9%, compared with 86.7% for Claude Code on Opus 5, the same source said.
Meta said it co-trained Muse Spark 1.2 with Muse Code so the core LLM and agent work together, and described Muse Spark 1.2 as a coding-focused upgrade that scaled up training compute on coding tasks and expanded training environment diversity. The launch also positions Meta as the latest tech giant to ship an agentic coding tool, competing with leading players like Anthropic and OpenAI, Decrypt added.