Today we're introducing Muse Ember, our smallest and most efficient model, built for complex reasoning execution and everyday agent work. Muse Ember can carry workflows across apps, call tools in sequence, run tests and make targeted code fixes. With a 1M-token context window and native understanding of text, images, audio and video, it can also work across large document sets and varied inputs.
Image from AI at Meta (@AIatMeta)
Image from AI at Meta (@AIatMeta)

Nov 9, 2026 · 6:04 PM UTC

271
354
304
3,319
1,271,218
RelevantRecentLikes
For agents that perform hundreds of small tasks, efficiency matters at every step. Muse Ember is built to deliver more intelligence for the cost. On the @ArtificialAnlys Intelligence Index, it achieves performance comparable to Claude Haiku 5.5 at roughly one-third of the cost per task, while scoring higher than GPT-6 Luna across all of Luna's reasoning settings.
Image from AI at Meta (@AIatMeta)
5
15
11
195
26,066
Muse Ember is available today on the Meta Model API, at $0.10 per million input tokens and $0.50 per million output tokens. Learn more on our research blog: go.meta.me/muse-ember
Introducing Our Smallest, Most Efficient Model: Muse Ember

Introducing Our Smallest, Most Efficient Model: Muse Ember

Muse Ember is our smallest and most efficient model, designed for complex reasoning execution and the work agents repeat all day.

research.meta.ai
8
11
3
154
18,938
packing native multimodal input with a 1M token context into a lightweight footprint is exactly what local developer tools need to run real agent loops without melting the gpu
1
1
20
2,617
Flexibility is everything 💪
9
2,417
Okay muse ember is really really fast.
Image from Deesha (@deesha_tech)
10
1,390
release the weights please. otherwise nobody cares
3
1
54
2,362
Naah you all cooked
13
2,371
can i download it or just rent it
25
2,671
Muse spark, Muse embora... The Nextel Will be Muse flamethrower
97
Wow
Image from RSI Guy (@RSIDataOTS)
5
1,526
ember's a great name for a small model. excited to try it
1
796
Don't care unless I see it's comparison with glm 5.3 flash.
Image from Avaaus (@avaausmamdani)
4
586
xitter.twitterapi.xyz/Marktechpost/status/21…
▲ Marktechpost AI (@Marktechpost) profile picture▲ Marktechpost AI
@MarktechpostNov 10
▲ Meta just released Muse Ember, its smallest model, and it matches Claude Haiku 5.5 on the Artificial Analysis Intelligence Index at about a third of the cost per task. The key idea is a split of labour inside agents. Muse Wonder plans the hard problem. Ember runs the volume: tool calls, tests, targeted code fixes, routing and extraction. It takes text, images, audio and video natively, with a 1M-token context and up to 512K output tokens. Ember scores 36 on the AA Intelligence Index at max effort, level with Claude Haiku 5.5 and ahead of GPT-6 Luna at 28. Meta's chart puts Ember at $0.136 per task versus $0.411 for Haiku. On AutomationBench-AA it completes 37.8% of workflows, versus 20.2% for Luna and 13.9% for Haiku. It is a trade-off though: Ember trails Haiku 5.5 on FrontierCode 2.0-AA, 27.2 vs 34.9, which judges whether real-repo changes are mergeable. Several results are also Meta internal runs. Available today on Meta Model API and Muse Code as muse-ember-1.0. $0.10 per 1M input tokens, $0.50 per 1M output. Weights are closed and parameter count is not disclosed...... Full analysis: marktechpost.com/2026/11/09/… Blog: research.meta.ai/blog/muse-e… Eval report: research.meta.ai/static/muse… API: developer.meta.com/ai/produc… @AIatMeta
Image from ▲ Marktechpost AI (@Marktechpost)
1
164
"cost per task" is the right number to look at, more than price per token. a cheap model that needs three retries can cost more than a pricier one that gets it right the first time. if the one-third figure holds up on real agent workloads, that's what would actually move people over.
6
1,798
Nice. I still can't unsee ALATMETA 😅
144
Sure but who’s dumb enough to trust their data to Meta?
1
129
@AIatMeta
1
1,726
will this be open sourced? @finkd
227
For the price, this model is really hard to beat.
62
Any performance data? (e.g., latency/throughput)
1
2
1,283
When will you release in South Africa? I would be more than happy to try your product, after many failed attempts with other agents.
608
Fix @AIatMeta as it has too many safeguards. I’ve cancelled my subscription. I couldn’t get it to even answer what model it was running. It kept lying. Then it was coy about the model it wa s running. Then all of a sudden it knew it was on Muse wonder. After saying it was on Muse Spark. But @Muse Agent. It knew immediately what model it was running. No ambiguity. It didn’t try to change the subject.
1,113
Smallest model carrying workflows across apps with 1M context is impressive
63
every house needs one small muse who actually gets things done. welcome, ember
906
Why not compare it to Deepseek V4.1 flash?
5
658
i mean, i won't mind if this gets available in open code ;)
488
umm.... My muse is called Ember from the beginning back in September, whats the deal here. lmao...
184
It's not your smallest model. Glimmer is your smolest model.
2
474
Maybe the AI is incredible. Maybe it isn't. Either way, Zuckerberg has spent years training me not to trust Zuckerberg. 😂👀 #AI #MetaAI
296
@grok what’s the user data usage and training policy for this meta model also drop pricing lil bro
1
660
If I can't use it locally then it's of no use for me ❌
276
hello Meta, please upload to huggingface, it is not there yet ;)
169
It's a good model. It doesn't have that obnoxious personality other models recently picked up, and it's got solid reasoning when deciphering a user's requests.
38
@grok can it run on my 32Go of RAM ?
1
418
Small models running full workflows is the future
12
Impressive to see such powerful agent capabilities packed into a smaller, more efficient model.
46
Load more replies