Muse Ember scores above Muse Wonder for under 5% of the cost
Published 2026-11-10 · Data captured 10 Nov 2026, 02:12 UTC
At high effort, Muse Ember scored 9.24/10 and ranked #36. A full run cost $0.156: 4.6% of Muse Wonder's best setting ($3.397, 9.20). No model scored higher for that money. Claude Haiku 5.5 and Particle-1 cost slightly less per solved test, and Gemini 3.7 Flash scored higher at about the same cost per solved test.
See current Muse Ember results →
- Available through Meta's Model API and OpenRouter; no public weights. Tested through OpenRouter at $0.10 input and $0.50 output per million tokens.
- Costs are recorded spend for one full run (23 tests, 3 repeats each). Cost per solved test is that spend in US cents divided by the solved tests.
Key results
Beats Muse Wonder for a fraction of the cost. 9.24 vs 9.20 for $0.156 vs $3.397, with 11.7 s vs 21.6 s average responses.
Set the effort yourself. The default, max, scored 8.64 for $0.346. High scored 9.24 for $0.156. Minimal scored 8.88 for $0.047 at 5.3 s.
Strong on agent work. 10.00 in agentic, coding, tool-calling and general-intelligence tests; 9.75 in instruction following, where Muse Wonder scored 6.55.
Weak on knowledge. Trivia 3.00, domain-specific questions 5.33, puzzles 5.41. Muse Wonder did better in trivia (4.67) and puzzles (7.67).
Strong rivals. Gemini 3.7 Flash low scored 9.37, solved 21 of 23 tests and answered in 6.2 s for $0.183. Claude Haiku 5.5 xhigh (0.773 cents) and Particle-1 low (0.836 cents) cost less per solved test than Ember high (0.865 cents).
| Model | Effort | Rank | Score | Solved tests | Correct attempts | Run cost | Cents per solved test | Avg response |
|---|---|---|---|---|---|---|---|---|
| Muse Ember | high | #36 | 9.24 | 18/23 | 55/69 | $0.156 | 0.865 | 11.7 s |
| Muse Ember | minimal | #67 | 8.88 | 17/23 | 53/69 | $0.047 | 0.276 | 5.3 s |
| Muse Wonder | medium | #39 | 9.20 | 17/23 | 53/69 | $3.397 | 19.982 | 21.6 s |
| Gemini 3.7 Flash | low | #30 | 9.37 | 21/23 | 66/69 | $0.183 | 0.868 | 6.2 s |
| Claude Haiku 5.5 | xhigh | #46 | 9.12 | 18/23 | 56/69 | $0.140 | 0.773 | 16.6 s |
| Particle-1 | low | #53 | 9.08 | 16/23 | 50/69 | $0.134 | 0.836 | 18.9 s |
| Gemini 3.6 Flash | high | #1 | 9.91 | 22/23 | 68/69 | $1.512 | 6.872 | 18.9 s |
Score and cost side by side
Gemini 3.6 Flash high leads at 9.91, but its run costs $1.512, about 10 times Ember high.
Score (out of 10, higher is better)
Full-run cost in USD (lower is better)
Average response time in seconds (lower is better)
Which effort to use
More effort did not help. High beat medium, xhigh and max on both score and cost; only minimal cost less per solved test. Muse Wonder also scored lower at high, xhigh and max.
Use high for quality and minimal for volume. Always set the effort: requests without one run at max.
| Model | Effort | Rank | Score | Solved tests | Correct attempts | Run cost | Cents per solved test | Avg response |
|---|---|---|---|---|---|---|---|---|
| Muse Ember | minimal | #67 | 8.88 | 17/23 | 53/69 | $0.047 | 0.276 | 5.3 s |
| Muse Ember | low | #86 | 8.64 | 15/23 | 50/69 | $0.114 | 0.757 | 8.2 s |
| Muse Ember | medium | #97 | 8.54 | 15/23 | 52/69 | $0.190 | 1.266 | 12.4 s |
| Muse Ember | high | #36 | 9.24 | 18/23 | 55/69 | $0.156 | 0.865 | 11.7 s |
| Muse Ember | xhigh | #73 | 8.75 | 16/23 | 53/69 | $0.232 | 1.450 | 13.9 s |
| Muse Ember | max | #87 | 8.64 | 15/23 | 50/69 | $0.346 | 2.304 | 19.0 s |
Strengths and weak points
Ember matched Muse Wonder in agentic, coding, tool-calling, data-extraction and general-intelligence tests, and led in instruction following and trick questions. It trailed in trivia and puzzles, so test knowledge-heavy tasks before you switch.
Muse Ember high
Muse Wonder medium
Who should choose it
Muse Ember high: agent steps, tool calls, coding subtasks and structured instructions at high volume.
Gemini 3.7 Flash low: when faster answers and more solved tests are worth a slightly higher cost.
Muse Wonder or a top-ranked model: trivia, specialist knowledge and puzzles.
Model details
| Context window | 1,048,576 tokens |
|---|---|
| Input | text, image, video, file, audio |
| Output | text |
| Price per 1M tokens | $0.100 input · $0.500 output · $0.004 cache read · $0.125 cache write |
| Reasoning | Always on; effort minimal, low, medium, high, xhigh, max (default max) |
| Parameters | Not disclosed by Meta. |
| Availability | API only (Meta Model API and OpenRouter); no public weights. |
| Prices observed | 2026-11-09 |
Method and limits
Every model ran the same 23 tests three times through OpenRouter. Scores are out of 10, ranks are leaderboard positions at capture time, and comparisons use the 408 complete, non-archived chat results.
Muse Ember and Particle-1 were tested in November 2026, the other models earlier. Tests 14, 15 and 22 kept the same prompts, answers and pass scores; only their recorded evaluator version differs.
Response times include provider and workload effects. Costs are recorded spend, not a price guarantee.
Sources
- Meta Muse Ember documentation (accessed 2026-11-09)
- Meta model list (accessed 2026-11-09)
- OpenRouter model page (accessed 2026-11-09)
Linked model pages show current results, which can differ from these figures.