Meta jumps into AI coding market in effort to chase Anthropic and OpenAI

Meta has taken a significant step in the artificial intelligence arena with the introduction of Muse Spark 1.1, a robust update to its first AI model, which debuted just three months ago under the guidance of AI chief Alexandr Wang. This new model aims to position Meta as a serious contender against established players like OpenAI and Anthropic. On Thursday, Wang discussed the advancements of Muse Spark 1.1 in an interview with CNBC, describing it as the company's most capable model for coding tasks to date. Initially, the first Muse Spark model was available only to select partners through a private API preview. However, Meta is now launching a public preview of the new model's API via a developer portal, allowing broader access for users interested in integration. According to a Meta representative, early partners have already been granted access to the API, while new users can join a waitlist for future access. Currently, API access is restricted to Meta's own platforms, with plans for expansion to third-party services still under consideration. This latest update follows another rollout earlier in the week, where Meta unveiled Muse Image, previously known as Mango. This model is designed for image creation, targeting creators and advertisers in an effort to broaden its user base. As Meta invests heavily in AI infrastructure, CEO Mark Zuckerberg faces increasing pressure from investors to demonstrate a return on this investment. While Meta's spending aligns with its hyperscaler competitors, it lacks a cloud infrastructure business and has struggled to keep pace with the development of popular AI models and applications. Wang emphasized that the pricing for Muse Spark 1.1 is competitive, with new API accounts starting off with $20 in free credits. Subsequent usage will be billed at $1.25 per million input tokens and $4.25 per million output tokens. "Our aim is to offer pricing that is appealing and scales effectively with usage," Wang stated. He also noted that Muse Spark 1.1 outperforms rival models in specific tasks related to coding and integration with various developer tools. The model was designed by Meta Superintelligence Labs to enhance AI agents’ capabilities, allowing them to autonomously tackle multiple tasks more effectively. The excitement surrounding AI agents surged in the first half of 2026, fueled by the popularity of OpenClaw, which enabled developers to manage AI models for advanced digital assistants. Wang shared that Muse Spark 1.1 has been trained to work seamlessly with popular tools used by developers, which he believes will encourage widespread adoption. While Meta previously focused on releasing its Llama models as open-source, the current strategy emphasizes monetizing access to proprietary models. Wang assured that Meta remains committed to open-source initiatives, revealing that a variant of Muse Spark is in development for future open-source release, although a timeline was not provided. Wang also shared his personal enthusiasm for using the latest Muse Spark model to enhance personal health management, such as web searching and accessing academic research. "It's a prime example of how these agentic systems can meet real-world needs," he concluded. In addition, he revealed that Meta is working on a more advanced AI model codenamed Watermelon, although details on its release remain under wraps.