Meta is jumping into the AI coding-agent race with Muse Code, a new terminal-based coding agent built to handle complete software engineering tasks across large repositories, putting the company in more direct competition with Anthropic’s Claude Code and OpenAI Codex.
The company released Muse Code in beta alongside Muse Spark 1.2, a coding-focused update to its AI model family. The agent can plan changes, write code, run validation, and keep working across long tasks that might span a large codebase.
“Releasing Muse Code in beta today. It’s a terminal coding agent that takes on complete software engineering tasks across large repos: planning changes, writing code, validating the results. Powered by Muse Spark 1.2, a coding-focused model update,” Meta CEO Mark Zuckerberg said in a post on X.
Releasing Muse Code in beta today. It’s a terminal coding agent that takes on complete software engineering tasks across large repos: planning changes, writing code, validating the results. Powered by Muse Spark 1.2, a coding-focused model update. pic.twitter.com/xqavk41w6v
— Mark Zuckerberg (@finkd) August 5, 2026
Meta AI chief Alexandr Wang described the tool in similar terms.
“You can install it with one command and then use it to take on complete software engineering tasks across a wide variety of use cases, planning changes, writing code, validating the results,” Wang said in a CNBC interview Wednesday.
Muse Code relies on persistent sub-agents that can maintain context and work in parallel, an approach aimed at longer software development jobs rather than one-off code generation.
Meta says Muse Spark 1.2 and Muse Code were developed and trained together. Wang said that pairing improves coding performance, though he declined to disclose usage figures for the Muse Spark models, saying “adoption has been exciting and strong.”
The model has posted competitive results on coding benchmarks including Terminal-Bench 2.1 and DeepSWE 1.1. It has a score of 54 on the Artificial Analysis Intelligence Index, placing it within striking distance of models from established AI rivals.
Meta Muse results on coding benchmarks (Credit: Meta)
Meta is betting on aggressive pricing to win developers
Raw capability is only part of Meta’s pitch. The company is using price as one of its biggest weapons against Anthropic and OpenAI.
Muse Spark 1.2 starts at $1.25 per million input tokens under Meta’s standard pay-as-you-go pricing. A contributor tier drops that price to $0.10 per million input tokens, more than 10 times cheaper, according to Wang.
That discount comes with a tradeoff.
Developers choosing the contributor tier must opt in to letting Meta use their data to improve its models. For companies unwilling to make that exchange, Wang said Meta has begun accepting requests for zero-data retention.
Wang called zero-data retention “a big enterprise feature that is important for folks.”
The distinction could matter for companies working with proprietary source code, customer information, or other sensitive material. Meta built one of the largest advertising businesses on the internet around data-driven targeting, so its handling of developer data is likely to receive close attention as Muse Code moves beyond beta.
Developers will be able to access Muse Code through the same Meta developer platform that hosts the Muse Spark API. Muse Spark 1.2 is set to appear on OpenRouter too, putting it alongside models from OpenAI, Anthropic, DeepSeek, Z.ai, and other AI labs.
Muse Code arrives as AI coding agents become one of the most contested parts of the AI market. Claude Code has gained traction among developers who want an AI system capable of working directly inside repositories and terminals. OpenAI has been pushing Codex deeper into software development workflows.
Meta now wants a seat at that table.
For Zuckerberg, the launch has a financial dimension. Meta continues to spend heavily on AI infrastructure and data centers, creating pressure to turn those investments into products developers and businesses will pay for. The company’s shares fell last week after Meta issued a lighter revenue forecast and reported shrinking free cash flow for the second quarter.
Muse Code gives Meta another path to monetize its AI research beyond consumer assistants and advertising.
The bigger question is whether developers will switch for price. Wang has made Meta’s strategy unusually clear: the company does not need Muse Code to crush Claude Code or Codex on every benchmark. If it can deliver comparable coding performance at a fraction of the cost, Meta may have a compelling way into a developer market where Anthropic and OpenAI already have a head start.
