AI agents ·
Meta's Muse Ember is the cheapest agent model yet — $0.10 per million tokens
Meta released Muse Ember on November 9, 2026, its smallest and most efficient model, built for the work AI agents repeat all day: tool calls, coding subtasks, routing, extraction, and long-document parsing. It is available now on the Meta Model API and in Muse Code as muse-ember-1.0.
The price is the story: $0.10 per million input tokens, $0.004 for cached input, and $0.50 per million output tokens. Meta says Ember matches Claude Haiku 5.5's best Artificial Analysis score — 36 — at about a third of the cost per task, roughly $0.14 per task against Haiku's $0.41. The catch is a price cliff: requests over 200,000 input tokens bill the whole request at triple rates, $0.30 input and $1.50 output.
Ember carries a 1-million-token context window and up to 512,000 output tokens, and it natively understands text, images, audio, and video. It is the first model Meta says it trained using Muse Wonder Full, its largest model, and Meta pitches the pair as a team: Wonder plans, Ember handles the volume. Ember is also available on Google Cloud and Oracle Cloud.
Why it matters
Agent bills are dominated by thousands of small repeated calls, not one big prompt — a model this cheap changes what it is economical to automate, from ticket triage to document extraction. The trade-off is the 200K price cliff: long agent loops that grow past it pay triple, so keep an eye on context size.
Sources
- Meta — Introducing Our Smallest, Most Efficient Model: Muse Ember
- eesel AI — Meta Muse Ember review (2026): benchmarks, pricing, and the catches
Spot an error? We correct quickly and note it. Contact us via the contact page.