Key Takeaways
- Meta opens Muse Spark 1.1 to developers through a public API preview.
- Competitive pricing targets OpenAI and Anthropic in the AI coding market.
- Meta shifts toward paid proprietary AI while maintaining open-source plans.
Meta Muse Spark 1.1 launched on Thursday as the company’s latest artificial intelligence coding model, opening public preview access through a developer portal as Meta seeks to compete with OpenAI and Anthropic while expanding its AI business through paid developer services.
Meta said Meta Muse Spark 1.1 is its strongest model for coding and AI agent tasks so far. The model follows the limited release of the original Muse Spark in April, which was available only to select partners through a private API preview.
Developers can now join a waitlist through Meta’s developer portal to gain API access over time. The company said the service currently runs only on Meta’s own infrastructure and is not yet available through third-party marketplaces.
“This is going to be served on top of the computer infrastructure that we’ve built,” Meta Chief AI Officer Alexandr Wang told CNBC.
Meta introduces competitive pricing for developers
The launch marks another step in Meta’s effort to diversify its revenue beyond advertising by charging developers for access to its AI models.
Wang described the pricing as competitive with offerings from OpenAI and Anthropic. New API accounts will receive $20 in free credits. After that, Meta will charge $1.25 per million input tokens and $4.25 per million output tokens.
“The goal is to really have attractive pricing that scales with immense consumption usage,” Wang said.
Earlier this week, Meta also introduced Muse Image, an AI image-generation model aimed at creators and advertisers.
The releases come as investors continue to watch whether Meta’s multibillion-dollar investments in AI infrastructure will generate stronger financial returns. While the company has increased spending on AI development, it continues to compete against established AI leaders, including OpenAI, Anthropic, and Google.
Company balances proprietary AI with open-source plans
Meta said Meta Muse Spark 1.1 was designed to perform coding tasks that help power AI agents capable of completing complex workflows with limited human involvement.
Wang said the model was trained to work with widely used developer tools and coding platforms to encourage adoption across the software industry.
“You kind of have to build coding capabilities as part of that in the service of overall agentic capabilities,” Wang said.
Although Meta previously emphasized open-source AI through its Llama models, the company is now placing greater focus on proprietary models available through paid access.
Wang said Meta remains committed to open-source development and confirmed that an open-source version of Muse Spark is in development, though he did not provide a release timeline.
He also said Meta is training a more advanced AI model, code-named Watermelon, but declined to say when it would become available.
Separately, Wang said he has been testing Muse Spark for personal health-related tasks, including searching academic research, browsing the web and organizing health information. He said such uses demonstrate the growing potential of AI agents to assist with everyday activities.
