AI News
All stories
Reported

Artificial Analysis says Meta's Muse Ember is the cheapest model scoring 36

Artificial Analysis says Muse Ember scores 36 on its Intelligence Index, matching Claude Haiku 5.5 (max) at about a third of the cost per task.

Models · · 2 min read

Artificial Analysis says Meta's Muse Ember (max) scores 36 on its Intelligence Index at $0.14 per task. That is the lowest cost of any model at that score. Claude Haiku 5.5 (max) reaches the same score at $0.41. The firm's benchmark measures cost against score on a Pareto frontier, which places Muse Ember at the cheap end of that curve. A lower-effort setting, Muse Ember (xhigh), scores 34 for $0.12. Artificial Analysis also says Muse Ember (max) is about four times cheaper than DeepSeek V4.1 Flash 1017 (max, scoring 33 at $0.60) and GLM-5.3-Flash (scoring 28 at $0.58).

The cost gap is partly a matter of how tokens are billed. Muse Ember has the same input and output token prices as Claude Haiku 5.5 and GPT-6 Luna: $0.10 input and $0.50 output per million tokens. Its cached input price is $0.004, which Artificial Analysis says is 60% cheaper than similarly priced models. The firm notes that cached input accounts for most of the cost in its agentic evaluations. Muse Ember also uses roughly a third as many output tokens as Claude Haiku 5.5 (max). Its Intelligence Index score is 8 points behind Muse Wonder (max), which scores 44 at $6.95 per task.

Agentic coding results are mixed. Muse Ember (max) scores 68.1% on AutomationBench-AA, level with Muse Wonder (max) and GPT-6 Astra, and beats Claude Haiku 5.5 (max) on GDP.pdf at 22.8% against 20.8%. It trails Haiku on AA-Briefcase, 1521 Elo against 1574. In coding, it beats Haiku 5.5 on Terminal-Bench 4.0 with 40.9% against 32.8%, but trails it on FrontierCode 2.0, 27.2% against 34.9%. When run in Muse Code, Meta's own coding agent, Muse Ember (max) scores 46 on the Artificial Analysis Coding Agent Index. That is ahead of Claude Haiku 5.5 in Claude Code (max, 37) and GPT-6 Luna in Codex (max, 31). The firm attributes that lead to SWE-Atlas-QnA-AA and Terminal-Bench 4.0.

Knowledge is the weakest area. On AA-Omniscience, Muse Ember's accuracy is 21.1%, compared with 34.7% for Claude Haiku 5.5 (max) and 50.5% for Gemini 3.8 Flash (high). Artificial Analysis says the gap is mostly because the model declines to answer 64% of questions. That same abstention keeps its hallucination rate at 17.4%, half that of Haiku 5.5. Readers comparing the two should weigh the accuracy and abstention figures together.

Muse Ember is available through Meta's first-party API and Muse Code. Artificial Analysis reports results at max effort, with xhigh also shown on its Intelligence versus Cost chart.

What changed

Type: Other · Applies to: Muse Ember (max), Muse Ember (xhigh), Claude Haiku 5.5 (max)

Why it matters

For agent workloads where cost per task drives the budget, Muse Ember offers a similar score at a fraction of the price of comparable models. Its low answer rate, though, means teams that rely on factual recall should expect more declined answers.

What you can do

Benchmark Muse Ember on your own agentic and knowledge tasks against your current low-cost model before switching, paying attention to cost per task and declined answers.

Original source

Where we saw it

1 source

Artificial Analysis

AI wrote this summary from the sources listed, and nobody reviewed it individually before publication. Check the original source before you rely on it. How it works