Hy4 vs Muse Ember
Muse Ember leads by 0.3 capability points (≈95% interval −2.3 to +2.9): statistically tied · P(higher) ≈ 59%
Wins 8 of 15 shared benchmarks (7 lost)
Capability vs. relative cost
Hy4Muse EmberClosed weightsOpen weights
Hy4 has no setting with a relative cost yet.
Benchmark wins7 – 8 · 15 shared
Comparing matching setups only (157 benchmarks excluded) · suites counted once
Area← Hy4Muse Ember →
Data: Surge AI, Mercor, DAXBench and 3 othersatlas.village.dev
Margins
Each benchmark scaled 0–100% (its declared range, else the field’s range); Δ is Muse Ember minus Hy4.
| FoodTruck Bench | Business simulations | 67598 | 34639 | −42.3 |
| Chartography (November 2026 rescoring) | Perception & media › Documents & charts | 49.2 | 79.3 | +30.1 |
| DAXBench v2 | Software › Data & ML | 87.2 | 62.8 | −24.5 |
| DAYJOB: Finance (October 2026 rescoring) | Professional › Finance, accounting & tax | 9.2 | 24.2 | +15.0 |
| Humanity's Last Exam Diamond | Knowledge & reasoning › General knowledge | 32.7 | 22.0 | −10.7 |
| DAYJOB: Healthcare (October 2026 rescoring) | Health & medicine | 7.2 | 17.2 | +10.0 |
| APEX-Agents 1.1 | Professional › General knowledge work | 85.1 | 75.7 | −9.4 |
| Harvey LAB (Mercor run, undisclosed ~213-task cohort) | Professional › Legal | 13.6 | 22.7 | +9.1 |
| ComplexConstraints (November 2026 rescoring) | Language & writing › Instruction following | 66.8 | 58.8 | −8.0 |
| EnterpriseBench: CoreCraft | Tool & computer use › Customer service & operations | 68.2 | 62.1 | −6.1 |
| HANDBOOK.md Agents v1.1 | Tool & computer use › Customer service & operations | 23.7 | 29.8 | +6.1 |
| GDP.pdf (November 2026 rescoring) | Professional › General knowledge work | 43.6 | 49.2 | +5.6 |
| τ³-Banking (Mercor run, bm25_grep retrieval) | Tool & computer use › Customer service & operations | 48.1 | 53.3 | +5.2 |
| RiemannBench (November 2026 rescoring) | Math | 51.2 | 53.6 | +2.4 |
| GDP.xlsx | Professional › General knowledge work | 17.7 | 15.4 | −2.3 |
Data: Surge AI, Mercor, DAXBench and 3 othersatlas.village.dev