Meta (Nasdaq: META) released Muse Spark 1.1 on Thursday, a coding and agentic AI model marking the company’s first-ever paid developer API, a sharp departure from the open-source Llama strategy that defined its AI approach until now.
The model is available through a new Meta Model API in public preview, priced at $1.25 per million input tokens and $4.25 per million output tokens. That puts it in a middle tier: costlier than OpenAI‘s cheapest option, GPT-5 Mini, and Anthropic‘s budget Claude Haiku 4.5, but cheaper than Anthropic’s flagship Claude Sonnet 4.6. New developer accounts get $20 in free credits to start.
“The pricing from some of the other labs is very extreme and has very high margins,” CEO Mark Zuckerberg told Bloomberg, adding that Meta believes it can offer “frontier or very high-level intelligence at a much more affordable cost.” By his own estimate, Meta’s rates run at roughly a quarter of what Anthropic and OpenAI charge for similar models.
Sponsored · Granada Gold Mine Inc.
Alexandr Wang, who leads Meta Superintelligence Labs, called it the lab’s “strongest model for agentic and coding work yet” in a CNBC interview. Meta says the model can write and debug code, operate software and external tools, understand text, images and video, and complete multi-step tasks with a 1-million-token context window, while delegating parts of complex tasks to sub-agents running in parallel.
The company claims Muse Spark 1.1 beats Google‘s latest Gemini release on coding and reasoning benchmarks, though it hasn’t published comparisons against OpenAI or Anthropic’s most current models.
The original Muse Spark, developed under the code name Avocado, launched in April through a closed partner program with no public API, and independent benchmarks showed it trailing well behind rivals specifically on coding and abstract reasoning tasks.
Muse Spark 1.1 arrives just two days after Meta released Muse Image, a separate model for generating images aimed at creators and advertisers. An open-source version of Muse Spark remains in development, Wang said, without offering a timeline. He has also flagged a more powerful model already in training, code-named Watermelon, but gave no release timeline for that either.
Zuckerberg also returned to posting on X for the first time in three years, specifically to announce Thursday’s launch.
