Meta Enters AI Coding Race With a Cheaper, Crash-Recovering Agent

On Wednesday, Meta officially entered the fast-growing market for AI coding tools. The company released Muse Code, its first software engineering agent, putting it in direct competition with products from OpenAI and Anthropic. Meta also confirmed that the agent will soon be available through third-party model platforms, which means developers will be able to reach it without leaving their usual tools.

Muse Code is built for long and complex work on large code bases. In one internal test, the agent ran for twenty-four hours straight and made more than one thousand tool calls while improving a system kernel. Unlike many earlier tools, it is designed to recover from a crash and continue from where it left off, instead of starting the whole job again. Meta says this makes it especially useful for engineers working on multi-day refactoring tasks.

Performance is solid, but not the very best. On a popular coding benchmark, Muse Code scored 82.9 percent, slightly behind Anthropic's top model but ahead of OpenAI's Codex and Grok Build. Meta's strategy is therefore not built on raw speed or top ranking. Instead, the company is competing on price and on practical reliability, betting that many teams will choose a cheaper and more dependable tool over the highest-scoring one.

Meta has also introduced a low "contributor" pricing tier that costs less than one tenth of the standard rate, on the condition that developers allow their data to be used for future model improvements. For larger customers, the company will offer a zero-data-retention option. With this mix of pricing, recovery features, and broad platform support, Meta is signalling that the next stage of the AI coding race will be fought on value and trust, not only on benchmark numbers.