Meta Platforms

AI Coding Tools · Developer Tools · Enterprise AI

Launch

Meta launches Muse Code beta to rival Claude Code and OpenAI Codex

The agent undercuts rivals mainly on price, not performance, trailing Claude and GPT on key coding benchmarks.

• Meta released Muse Code, a beta terminal coding agent powered by its new Muse Spark 1.2 model, aimed squarely at Anthropic's Claude Code and OpenAI's Codex.
• Built by Meta Superintelligence Labs under AI chief Alexandr Wang, Muse Code plans, writes, and validates code across large repositories using coordinated persistent subagents.
• Meta is competing on cost rather than raw capability: a discounted 'contributor tier' that lets Meta train on user data is over 10x cheaper than the standard pay-as-you-go pricing.
• Muse Spark 1.2 scored 82.9% pass@1 on Terminal-Bench 2.1, edging past GPT-5.6 Terra but behind Claude Opus 5, and lagged both rivals on the DeepSWE 1.1 benchmark at 59.3%.
• The launch follows a sharp drop in Meta's stock after a light revenue forecast and shrinking free cash flow, underscoring pressure to show returns on its AI infrastructure spending.
• Meta is betting that pricing pressure, not benchmark superiority, can carve out developer share in a coding-agent market Anthropic and OpenAI currently dominate on capability.