Meta Platforms Inc. rolled out a major update to its artificial intelligence (AI) lineup on Thursday: Muse Spark 1.1 challenges OpenAI and Anthropic in the high-stakes market for paid enterprise and developer tools.
Optimized for advanced coding and autonomous agentic workflows, Muse Spark 1.1 represents Meta’s most aggressive push yet to close the gap with its hyperscaler peers. The model can write and debug code, manage complex multi-step digital workflows, and orchestrate tasks across third-party software applications with minimal human intervention.
The launch represents a significant departure from Meta’s focus on its open-source Llama family of models. Instead, the company is prioritizing a proprietary, pay-as-you-go developer ecosystem.
Alexandr Wang, head of Meta Superintelligence Labs (MSL), said Muse Spark 1.1 was specifically trained on code-heavy tasks to supercharge its digital assistant capabilities. Industry excitement surrounding autonomous AI agents spiked earlier this year with the popularity of frameworks like OpenClaw, and Meta has ensured Spark 1.1 integrates seamlessly with these widely used developer harnesses.
The rollout follows immense pressure from Wall Street for Meta to demonstrate financial returns on its massive capital expenditures in AI infrastructure. Unlike Microsoft Corp., Google, or Amazon.com Inc., Meta currently lacks a core cloud infrastructure business to monetize its research, making direct developer APIs a critical revenue bridge.
To drive rapid adoption, Meta is introducing what Wang described as “very aggressive and attractive” pricing. Input costs $1.25 per million tokens. Output is $4.25 per million tokens. Users receive $20 in free credits for new developer API accounts.
This structure positions Muse Spark 1.1 below premium models like Anthropic’s Claude Sonnet, though slightly above lower-tier options like OpenAI’s GPT-5 mini.
The release was notable enough to prompt Meta CEO Mark Zuckerberg to post on the social media platform X for the first time in three years. Zuckerberg touted the model’s “exceptional performance in personal agentic tasks” and its low-cost structure.
Muse Spark 1.1 is already available in “Thinking mode” within the Meta AI app and website. Meta eventually plans to use the technology to replace older Llama models powering chatbots across WhatsApp, Instagram, Facebook, and its Ray-Ban smart glasses.
The launch caps a busy week for Meta’s AI division, which also unveiled its first native image-generation model, Muse Image (code-named Mango), on Tuesday. Despite the commercial focus, Wang noted that Meta remains “committed to open source” and is currently developing a specialized open variant of the Muse model.
Looking ahead, Meta is already training an even more powerful foundation model code-named Watermelon, signaling that the company has no plans to slow down in the intensifying AI arms race.
