Meta AI has launched Muse Ember, which Simon Willison covers in a link post on his weblog. Meta describes it as its most compact and efficient model, designed for the routine jobs agents perform constantly: tool calls, coding subtasks, routing, extraction, and parsing of long documents. Willison places it in the same class as Claude Haiku and GPT Luna.
Pricing
For shorter prompts, Ember costs the same as GPT-6 Luna and Claude Haiku 5.5: $0.10 per million input tokens and $0.50 per million output tokens. Willison notes that Claude Fable 5.2, released the same day, costs $8 input and $40 output, which is 80 times as much.
All three small models accept up to a million input tokens, and each raises its price beyond a threshold. The thresholds differ:
- Muse Ember 1.0: up to 200k tokens, $0.10 input, $0.004 cached input, $0.50 output. Above 200k, $0.30 input, $0.012 cached, $1.50 output.
- GPT-6 Luna: up to 272k tokens, $0.10, $0.01 cached, $0.50 output. Above that, $0.20, $0.02 and $0.75.
- Claude Haiku 5.5: up to 100k tokens, $0.10, $0.01 cached, $0.50 output. Above that, $0.50, $0.05 and $2.50.
Once the threshold is counted, Ember lands between the other two. Its cached-input rate is the lowest of the three at short length.
Price per token is not the whole cost
Willison points out that reasoning tokens mean two models can use very different numbers of tokens on the same problem. He therefore leans on the Artificial Analysis chart of Intelligence Index against cost per task. On that chart Ember sits on the Pareto line, where improving one dimension means giving up the other. Meta also claims in its announcement that no model on the full leaderboard scores higher for less cost per task. That is Meta's claim, relayed by Willison from the chart.
Reasoning levels
Ember offers six reasoning levels: minimal, low, medium, high, xhigh and max. Willison ran his pelican-on-a-bicycle SVG test at each level. The "high" level did not return a complete SVG, and image quality did not rise cleanly with reasoning level. He found the results comparable to those from Muse Wonder, which costs 40 times as much as Ember. This is an informal test and says little about business tasks.
What the source does not cover
The post does not say where Ember is available or how to access it, so check Meta's own announcement before planning a migration.