Clarity for AI buildersSponsored by AI Routing
pricing

Meta introduces Muse Ember for routine agent tasks at lower reported costs

Meta introduces Muse Ember for routine agent tasks at lower reported costs
Meta has introduced Muse Ember, a model for routine AI agent work, according to The Decoder. The report describes Muse Ember as Meta’s smallest and most efficient AI model to date, aimed at tool calls, small coding tasks, request routing, data extraction and long-document processing. It is available through the Meta Model API and Muse Code, with pricing of $0.10 per million input tokens and $0.50 per million output tokens. The reported benchmark results suggest a substantial cost difference within Meta’s lineup. Muse Ember scored 36 on the Artificial Analysis Intelligence Index at about $0.14 per task, compared with a score of 37 and a cost of $2.39 per task for the larger Muse Spark 1.3. On that measure, Ember delivers nearly the same score at a much lower reported task cost. That comparison does not establish equivalent performance on every workload, but it supports the model’s positioning for repetitive tasks where cost matters. According to The Decoder, Meta also claims that Muse Ember delivers more intelligence per dollar at every reasoning level than Claude Haiku 5.5 or GPT-6 Luna. The report says Ember leads in most of Meta’s own tests, but identifies a coding exception: on real coding tasks in FrontierCode 2.0-AA, Haiku scored 34.9 percent against Ember’s 27.2 percent. That result limits any broad reading of Ember’s cost-efficiency claims as evidence of superior coding ability. Muse Ember supports text, images, audio and video, and can process up to one million tokens at once, the report says. Those capabilities extend its stated scope beyond short text requests to multimodal inputs and long documents. The context capacity alone, however, does not demonstrate how accurately it handles every part of a long input. For developers considering the model, the evidence points to a trade-off rather than a universal replacement for larger models: low reported task costs and a competitive aggregate score, alongside a weaker result than Haiku on the cited coding benchmark. The per-task benchmark costs and the API’s per-token prices are different measures; the supplied report does not provide enough workload detail to translate one directly into the other.