Meta enters the developer battle: launches new code agent to compete with Anthropic and OpenAI
Meta has introduced Muse Code, an autonomous programming agent designed to compete with tools from Anthropic and OpenAI. The tool enables developers to automate code planning, writing, and testing.
Meta is entering the arena of AI-based programming tools and launching Muse Code — a new code agent designed to compete with tools from leading AI companies like Anthropic and OpenAI. The new tool allows developers to use an autonomous agent to perform software engineering tasks, ranging from planning code changes to writing, testing, and verifying results.
Muse Code is one of the first significant launches under the leadership of Alexander Wong, head of Meta's AI division and leader of Meta Superintelligence Labs, which is responsible for developing the company's base models. Wong joined Meta last year as part of CEO Mark Zuckerberg's move to strengthen the AI field following criticism of the company's pace of progress compared to its competitors.
"You can install it with a single command, and then use it to perform full software engineering tasks in a wide range of use cases — to plan changes, write code, and verify the results," Wong said in an interview.
The launch of Muse Code comes against the backdrop of the growing struggle between tech giants for control over the market for AI tools for developers. Meta's tool joins products like Anthropic's Claude Code and OpenAI's Codex, which allow developers to build applications and manage development processes through an interface based on AI agents. The new tool is currently available in a preview version and operates alongside the company's latest AI model, Muse Spark 1.2.
Wong did not provide usage data for the Muse Spark models, but said that "adoption has been impressive and strong." According to him, the latest Muse Spark model was developed in parallel with Muse Code, which helped improve the system's coding performance.
Meta is choosing to compete primarily through price. Developers will be able to use Muse Code on a pay-per-use model, alongside a cheaper plan intended for contributors. Wong claimed that this plan would be "more than 10 times cheaper even than the pay-per-use level." However, users on the cheap plan will have to agree to Meta using their data to improve the model.
According to Wong, the company is not necessarily trying to beat Anthropic and OpenAI in capabilities, but rather to differentiate its products through lower cost. "We are differentiating the new coding tool and the Muse Spark family of models through price," he said.
At the core of Muse Code is a mechanism called harness — a wrapper that allows developers to manage AI models adapted for different software projects. The service will be available through Meta's developer page, where the company also offers access to the Muse Spark model's API, as well as through the OpenRouter platform, which hosts a variety of AI models.
Meta has also started accepting requests for a zero-data retention service, where developer data will not be saved for model improvement. "This is a significant enterprise capability that is important to many users," Wong said.
The move comes at a time when Meta continues to pour huge investments into AI infrastructure, including data centers and computing systems. The company is trying to turn these investments into a new growth engine, even though currently about 98% of its revenue still comes from online advertising based on user data analysis.
