Meta's Muse Spark 1.1 Takes Aim at OpenAI and Anthropic in AI Coding Race

Meta is making a serious push into the competitive AI coding market, rolling out a major update to its Muse Spark model just three months after the initial launch. The new version, Muse Spark 1.1, is being called the company's "strongest model for agentic and coding work yet" by AI chief Alexandr Wang. The update arrives as Meta tries to carve out a stronger position against established players like OpenAI and Anthropic.

Unlike the first Muse Spark, which was limited to select partners via a private preview, the new model's API is now available through a public developer portal. Users can sign up and get integration instructions, though access is currently limited to Meta's own platforms rather than third-party marketplaces like OpenRouter. Early partners already have access, and new users can join a waitlist. Wang noted the pricing is "very aggressive," with new accounts receiving $20 in free credits and rates set at $1.25 per million input tokens and $4.25 per million output tokens.

The move is part of a broader strategy to show returns on Meta's massive AI investments, even as the company lacks a cloud infrastructure business like its hyperscaler rivals. Wang said Muse Spark 1.1 outperforms competitor models on certain coding tasks, which is key for building AI agents that can work autonomously. Meta is also training a more powerful model code-named Watermelon, though a release date hasn't been set. While the company has shifted focus to selling proprietary models, Wang said a future open-source variant of Muse Spark is in development.

Source: CNBC