The company that made open weights mainstream now competes on discounts
Meta's new coding agent undercuts the market at 20¢ per million tokens—but there's a catch.
Why it matters
Meta is shifting from open-weight leadership to competing on price and developer experience (with Muse Code, a crash-resilient coding agent). The data-sharing tier reveals a new tradeoff in the commoditizing LLM market: cheaper access for training consent.
The key facts
12 to knowMeta released Muse Spark 1.2 + Muse Code (coding agent)
Cheapest tier: $0.20 per million output tokens
Requires data sharing for training on lowest tier
Muse Code designed to resume after crashes
Meta competing on price, not top-end performance
Benchmarks have a 'glaring gap' (unspecified)
Meta released Muse Spark 1.2 with Muse Code coding agent
Cheapest tier priced at $0.20 per million output tokens
Requires data sharing for training on lowest-cost tier
Agent feature: persistent state across crashes
Strategy: competing on price and UX, not top-end performance
Noted benchmark gap in the model
Go to the source
The Decoderthe-decoder.com
Publisher excerpt: Meta released Muse Spark 1.2 along with its own coding agent, Muse Code, which is designed to pick up exactly where it left off after a crash. The cheapest tier runs just 20 cents per million output tokens but requires users to share their data for training. Meta is competing on price, not top-end…