1Meta Muse Ember
No. 1 — Meta Muse Ember. On the lists of 8 of 10 AIs; median place there: 1.5. The AIs put it first because it finishes far more office workflow steps than Haiku, even though it loses on merge-ready code.
Did you know
Meta Muse Ember is the smallest model in Meta's Muse family. It is meant to run alongside a bigger planner, Muse Wonder. Meta's line: "Muse Wonder plans, Muse Ember handles the volume." Its weights are kept private.
- What it is
- Meta's smallest, cheapest Muse model, built for small, repeated agent jobs.
- Its edge
- It finishes 68.1% of office workflow objectives, well above Haiku's 35.4%.
- Who it suits
- Suits office-style agent work; not for code that must be merged.
- Office workflow objectives completed68.1% of office objectives
Why it is here
On Meta's long desktop-workflow test (OSWorld 2.1), Ember scores 69.5% partial credit. Claude Haiku 5.5 scores 68.9%. That is nearly a tie, so Ember's office-work lead is small on this measure.
On fully finished tasks, Ember completes 33.8% against Haiku's 27.8%, per The New Stack. Its edge shows up when you count only tasks done start to finish.
Did you know
Ember costs $0.10 per million input tokens and $0.50 per million output tokens. It still has a 1M-token context window and takes image inputs. It is the fourth Muse model, after Spark, Glimmer and Wonder.
Keep in mind
Ember launched on November 9, 2026 as a smaller, cheaper model, so coverage so far leans on Meta's own benchmarks. I found no independent user reviews yet. Treat the early scores as the maker's claims.
Where to go further
Open the blog and try Muse Ember from the 'Try Muse' button. Look first at the 'Intelligence vs. Cost Per Task' chart right after the intro. These numbers are Meta's own, so check them against independent tests.