Meta launches metered Muse Spark 1.1 API priced up to 83% below Claude

Meta introduced Muse Spark 1.1, a multimodal reasoning model aimed at coding and agentic workloads, priced as much as 83% below Anthropic's Claude. The most consequential detail isn't the price — it's the business model: Muse Spark 1.1 is Meta's first proprietary, metered API rather than a free open-weights release, marking a sharp departure from the open Llama strategy that defined Meta's AI identity.

Strategically, the move signals Meta chasing direct API revenue and tighter control over a frontier model, betting that aggressive pricing can win developers who might otherwise default to OpenAI, Anthropic, or cheap Chinese challengers like DeepSeek. Undercutting Claude by up to 83% while offering multimodal reasoning positions Muse Spark as a value play in the agentic-coding segment, where cost per task compounds quickly at production scale.

But the pivot is culturally fraught. Meta built enormous developer goodwill by releasing Llama weights openly, and shifting a flagship model behind a paywall risks the same 'betrayal' backlash that erupted in prior open-source U-turns. The AI community has historically rewarded Meta specifically for openness, so a proprietary metered model invites accusations of abandoning the ethos that differentiated it from OpenAI and Google.

The competitive context is a market bifurcating between commoditized cheap inference (DeepSeek, MiniMax) and premium closed models (Anthropic, OpenAI). Meta is attempting to straddle both — undercutting premium players on price while closing the weights like they do. Whether developers accept a metered Meta model, or view it as a broken promise, will shape adoption. Watch benchmark comparisons against Claude and GPT-5.6 for coding, and whether Meta continues any open Llama releases alongside the new paid tier.