The Muse Spark 1.1 model is priced at $1.25 per million input tokens and $4.25 per million output tokens, with $20 in free credits to start
SOPA Images / Getty Images
Meta $META launched a public developer API for its Muse Spark 1.1 model on Thursday, marking the first time the company has charged businesses for access to one of its AI models. The move positions Meta against Anthropic and OpenAI in the market for AI developer tools.
The new model's API is priced at $1.25 per million input tokens and $4.25 per million output tokens, the company said. New accounts will receive $20 in free credits. Meta CEO Mark Zuckerberg characterized that pricing as roughly 25% of what Anthropic and OpenAI charge for comparable models, according to Bloomberg. Alexandr Wang, who leads Meta Superintelligence Labs, described the pricing as "very aggressive and attractive" in an interview with CNBC.
Access is being offered through a developer portal currently in public preview. Some early partners have already been onboarded, while others can request to be added via a waitlist. Meta has said it will not make the model available through outside platforms such as OpenRouter for the time being, keeping distribution confined to its own properties.
Built to handle agentic and coding work, the model targets use cases where AI completes a series of tasks autonomously on someone's behalf. It supports a context window of one million tokens and is capable of running multiple subagents simultaneously, Meta said. Zuckerberg described the model's agentic reasoning and tool use as "state-of-the-art or very close to it."
According to Wang, strong coding ability is a prerequisite for the kind of autonomous, multi-step behavior that defines agentic AI — meaning the two capabilities are deeply intertwined. Zuckerberg said the model outperformed Google $GOOGL's Gemini across a range of benchmarks tied to agents, coding, and multimodal tasks, and said Meta staff have already been using it to develop new features across its apps.
Early access partners include Replit, Cline, and Box, the company said.
The launch comes as Meta moved away from its open-source Llama model strategy in April with the debut of the original Muse Spark — a proprietary model available only to select partners via a private API preview. Coding was identified as a relative weakness of that initial release. An open-source version of the model is in development, Wang said, though he gave no indication of when it might be released.
A next-generation model is in training at Meta under the internal codename Watermelon; both Zuckerberg and Wang acknowledged its existence but offered no release timeframe. The 1.1 model being released Thursday was itself developed under the codename Avocado.
Thursday's API announcement follows Meta's release of Muse Image on Tuesday, a model for generating images that is available to consumers and advertisers. The two launches are part of Meta's effort to demonstrate returns on its AI investment as Zuckerberg faces pressure from Wall Street over the company's growing infrastructure spending.
Join 500,000+ readers who start their day with Quartz.
By subscribing, you agree to our Terms of Service and Privacy Policy.
