whatmodel.sh / Muse Ember

Muse Ember is a closed-weights language model from Meta, released November 2026. Licence: proprietary.

Catalogue identity notes

Announced 2026-11-09 by Meta ('Introducing Our Smallest, Most Efficient Model: Muse Ember'); generally available the same day on Meta Model API and Muse Code as muse-ember-1.0. Distinct from Muse Spark and Muse Wonder (Meta: Muse Ember is the first model trained using Muse Wonder Full). Parameters not published. Arena entry 'Muse Ember (Max)' (webdev, org 'meta', Proprietary) is this model at max reasoning effort; Meta's own evaluations also use max effort. Mapped like 'Muse Wonder (Max)' -> muse-wonder.

Model information sources

Release date ↗

75.6±1.3#31 of 743 LLMsfrom 3 benchmark results across 2 sources

Score within the LMArena (text) comparison set only.

$ how it compares

The LLMs ranked either side of Muse Ember, out of 897 with a score, then more from Meta.

#modelscorewhereRAMreleased
29GPT-5.2 Pro OpenAI76±1api $57.75/M2025-12
30Claude Haiku 5.5 Anthropic76±1api $0.20/M2026-10
31Muse Ember Meta76±1api $0.20/M2026-11
32GPT-5.4 OpenAI75±1api $5.63/M2026-03
33DeepSeek V4.1 Flash DeepSeek75±1local~512GB2026-09
more from Meta
10Muse Wonder Full Meta81±1Muse Code only2026-10
20Muse Wonder Meta78±1api $8.00/M2026-10
34Muse Spark 1.3 Meta75±1api $2.00/M2026-09

see all LLMs · find the ones that fit your machine

$ what it's good at

agent tasks▁▁▃▆█▄▅▅▃▃▂▁ #29 of 319 * · 67% of its score · 2 tests

Beats GPT-5.5, GPT-5.6 Terra, Hy4 preview (770B-A49B) here, though they rank higher overall. Trails GLM-5.3-Flash, GLM-5.3, DeepSeek V4.1 Flash here, though they rank lower overall.

testresultvs the others in its rundifficultycountswhy
Arena Agent (October 2026 aggregate IPS)0.0167 IPS ✎▁▁▁▃▃█▂▁▁▁·▁ beats 36/63avg -0.403 IPS▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 54%sharp
Terminal-Bench 4.130.6% ✎▂▂▂··▅·▂▅█▃▆ beats 10/20very hard · avg 1.0%▇▇▇▇▇ 12%sharp · at its level · noisy
coding▁▁▁▃▆▇█▆▆▃▃▁ #33 of 474 * · 33% of its score · 1 test

Beats GPT-5.5, GPT-5.6 Terra, Claude Opus 4.8 here, though they rank higher overall. Trails Qwen3.8-Flash-Next, Grok 4.7, Qwen3.8 Max (0902) here, though they rank lower overall.

testresultvs the others in its rundifficultycountswhy
Arena WebDev (Code Arena)1608.8 ✎▁▂▄▃▇█▅▇▇▅▂▁ beats 101/134avg 996▇▇▇▇▇▇▇▇▇▇▇▇▇ 33%sharp

$ where to run it

Through an API

api · $0.20/M · $0.1 in / $0.5 out per 1M tokens · 1M context · dev.meta.ai ↗ · price sourceUSD per 1M tokens, Meta Model API Standard mode for muse-ember-1.0 (no Contributor-mode variant), requests up to 200K input tokens; cached input $0.004 excluded. Requests over 200K input tokens are billed in full at $0.30 in / $0.012 cached / $1.50 out. Same figures on Meta's own pricing page https://dev.meta.ai/docs/pricing-rate-limits#muse-ember-pricing and in the announcement. Observed 2026-11-12.

$ how the score was worked out · LMArena leaderboard dataset 87%, Terminal-Bench 4.0 official l… 12%

Each result implies a score on its own, after adjusting for how hard its test is and who ran it. The score is their weighted average, with a small pull toward the middle for thinly tested models. Each test's weight is its counts column above; here it's added up by who ran the tests. How it works

sourceresultsshare of the scorelicence
LMArena leaderboard dataset2CC BY 4.0
Terminal-Bench 4.0 official leaderboard submissions1Apache-2.0
prior · pull toward the middle
weighted average= 75.6

$ improve this page

Use a short GitHub form or edit the data through a pull request. Every number needs a source; here's which ones we can use.

data updated · sources and licences · json