Meta Unveils its First AI Coding Agent to Compete with OpenAI and Anthropic

With the release of Muse Code, Meta's first coding agent, the company is aiming to compete with top AI players such as Athropic and OpenAI. Alexandr Wang, head of AI at Meta Superintelligence Labs and supervisor of foundation model development, has just released his most recent important work, Muse Code.

Wang became a part of Facebook in June of last year and has since been at the center of CEO Mark Zuckerberg's plan to fix the company's AI strategy. Users can install Muse Code with just one command, according to Wang. After that, users can use it to plan modifications, write code, and validate outcomes, among other software engineering duties, for a broad range of use cases.

Benefits of Meta’s Muse Code

Zuckerberg plans to monetise AI in several ways, and the new coding agent is just one of them. Data centres and associated computing infrastructure continue to receive substantial investments from Zuckerberg's company. Similar to Claude from Anthropic and the Codex helpers from OpenAI, the new tool streamlines the app development process under a unified interface.

Nevertheless, it goes beyond that by overseeing teams of digital agents powered by AI, which may support software development. The most recent artificial intelligence model from the firm, Muse Spark 1.2, is compatible with Muse Code, which is now available in preview. Wang was tight-lipped about the company's Muse Spark AI models' user numbers. But he did say that adoption has been robust and exciting. Wang claims that the new Muse Spark model, which was built and trained in tandem with Muse Code, enhances the coding performance as a whole.

Muse Code offers a pay-as-you-go alternative for developers. According to Wang, the agency offers a contributor tier that allows users to access the platform at a much reduced fee. According to him, the low-cost tier is over ten times less expensive than the pay-as-you-go tier. When asked about Meta's usage of third-party data to strengthen the underlying technology, Wang responded that developers under the cheapest tier were required to opt-in in order to help enhance the model.

Meta’s Marketing Strategy for Muse Code

Wang added that in comparison to popular products like Anthropic and OpenAI, Meta is differentiating its new AI coding tool and Muse Spark family of models based on price rather than capabilities. A so-called harness is the backbone of Muse Code; it allows developers to handle AI models that are specifically designed for coding projects.

Muse Spark, the company's artificial intelligence model API, and Muse Code, according to Wang, will both be available for purchase on the same Meta developer page. The OpenRouter platform, which hosts popular AI models, will also have Meta's more recent model available. Among these AI models are DeepSeek and Z.ai, two so-called open-weight AI models developed in Chinese labs. In addition, according to Wang, Meta is beginning to take requests for zero-data retention, which means the business will not save developer data in order to enhance models.

It stands for a major enterprise aspect that people care about, he said. Online advertising, which Meta controls by using user data to target consumers, accounts for 98% of the company's revenue.