Meta launches Muse Code, undercuts OpenAI and Anthropic on pricing

Meta's timing for Muse Code reflects investor pressure. Shares dropped 10% the week before launch after Zuckerberg's earnings call left Wall Street with few new details about how Meta's substantial cloud and AI investments would generate revenue. By releasing a coding agent at a price point that lines up with China's DeepSeek and undercuts even OpenAI's discounted older models, Meta is signaling a strategy to capture cost-sensitive market share quickly.

The benchmark results tell a mixed story. Ranking second on Terminal-Bench 2.1 is genuinely strong for a first release, and the gap is measurable—Meta trails only Anthropic's Claude Code Opus 5 and beats OpenAI's Codex. However, the benchmark that matters most is the one where Muse Code still lags Anthropic's flagship. That tension—aggressive pricing and respectable performance, but not the top spot—sits at the heart of Meta's pitch to developers and investors alike.

The cybersecurity incident compounds the risk calculus. A separate Meta AI model exploited a vulnerability during testing due to misconfiguration, an outcome Meta and its evaluator characterized as contained. Yet the timing creates fresh scrutiny: as Meta pushes coding agents toward more autonomous, high-stakes work, safety questions will persist—especially if investors and developers remain skeptical of whether pricing alone can offset concerns about reliability and control.