AI
Meta launches Muse Spark 1.1 with paid API, undercutting OpenAI and Anthropic on price
Meta's newest AI model emphasizes agentic capabilities and marks the company's first paid frontier model API, with pricing at roughly a quarter of competitors' rates.
The New Stack
Meta has rolled out Muse Spark 1.1, a significant refresh of its frontier AI platform just three months after the initial release. The launch arrives as CEO Mark Zuckerberg accelerates the company's effort to narrow the gap with leading AI labs like OpenAI and Anthropic.
The new model comes with a meaningful strategic shift: Meta is introducing its first paid API tier for a frontier model, allowing developers to access the system through Meta's own hosted infrastructure rather than self-hosting or relying on third-party providers. Developers can claim $20 in free credits initially, then face charges of $1.25 per million input tokens and $4.25 per million output tokens — roughly 75% cheaper than the premium offerings from rivals.
A new business model#
Until now, Meta has primarily distributed its Llama models as open-weight releases that developers download and host themselves. The Muse Spark 1.1 API fundamentally changes that approach, positioning Meta as a direct competitor in the managed inference space dominated by OpenAI, Anthropic, and Google.
Zuckerberg framed the price strategy candidly: "The pricing from some of the other labs is very extreme and has very high margins. We think that there's a real ability to be able to offer frontier or very high-level intelligence at a much more affordable cost."
This represents a notable pivot in Meta's enterprise strategy. Rather than betting solely on open models to build developer goodwill, the company is now actively capturing recurring API revenue while keeping developers within its ecosystem. Meta plans to offer this model through a public waitlist on its developer portal, though it won't be distributed through third-party marketplaces initially.
Agentic AI as the focus#
Muse Spark 1.1 prioritizes agentic capabilities — the ability for AI systems to autonomously complete multi-step tasks, call external APIs, and coordinate workflows. Under Alexandr Wang's leadership at Meta Superintelligence Labs, the team invested heavily in coding fluency, reasoning that strong code generation underpins the tool-use and reasoning skills AI agents require.
Wang described the model as possessing "state-of-the-art or very close to it" agentic reasoning, with particular strength in how it handles third-party developer tools and integrates with existing workflows. Zuckerberg added that internal benchmarks show Muse Spark 1.1 outperforming Google's Gemini on agentic and coding tasks — a claim he framed as historically significant for Meta's AI competitiveness.
