The floor just dropped. Meta launched Muse Spark 1.1 through an API priced at roughly a quarter of what competitors charge. For developers running coding agents where every token gets billed, that’s not a minor discount — it’s a new default option for the high-volume, low-judgment calls that make up most of a session.
The headline number is clear: 25% of the field’s going rate. That makes Muse Spark 1.1 the cheapest credible fallback in any orchestrator’s routing table right now. Whether it holds quality on complex tasks is the open question — but for the boring 80% of lookups, linting, and simple generation, the pricing alone earns it a slot.
For terminalblog’s audience, this changes the math on multi-model routing. If you were already composing models — letting a router pick the cheapest adequate model per subtask — Muse Spark 1.1 slots in as the cost-conscious backbone. The vendors that win the pricing race early don’t need to win every benchmark; they just need to be cheap enough that the router picks them first.
One line: Muse Spark 1.1 doesn’t need to be the best — at a quarter of the price, it just needs to be good enough to own the cheap slot in every router.