GPT-6 Astra vs Muse Ember
TryTop 2 now
GPT-6 Astra leads by 4.9 capability points (≈95% interval +2.4 to +7.5): a clear lead over Muse Ember · P(higher) ≈ 100%
Wins 40 of 57 shared benchmarks (17 lost)
Capability vs. relative cost
GPT-6 AstraMuse EmberClosed weightsOpen weights
Benchmark wins40 – 17 · 57 shared
Comparing matching setups only (9 benchmarks excluded) · suites counted once
Area← GPT-6 AstraMuse Ember →
Data: Artificial Analysis, Surge AI, Vals AI and 12 othersatlas.village.dev
Margins
Each benchmark scaled 0–100% (its declared range, else the field’s range); Δ is Muse Ember minus GPT-6 Astra.
| TaxCalcBench TY25 | Professional › Finance, accounting & tax | 46.0 | 2.0 | −44.0 |
| Humanity's Last Exam Diamond | Knowledge & reasoning › General knowledge | 63.8 | 25.1 | −38.7 |
| Terminal-Bench-Science 0.1 | Science › Multidisciplinary | 54.8 | 17.6 | −37.1 |
| DAXBench v2 | Software › Data & ML | 96.8 | 62.8 | −34.0 |
| Code Migration v1.1 (Vals Index v2.2 subset) | Software › Software engineering | 78.0 | 47.1 | −30.9 |
| KradleBench | Games | 56.1 | 25.7 | −30.4 |
| FoodTruck Bench | Business simulations | 57761 | 34639 | −29.7 |
| RiemannBench (November 2026 rescoring) | Math | 81.6 | 53.6 | −28.0 |
| LMArena Agent Arena Net Improvement (Steering Burden) | Tool & computer use › General agents | 14.6 | 1.0 | −26.5 |
| Terminal-Bench 4.0 (Artificial Analysis) | Software › Software engineering | 59.6 | 35.4 | −24.2 |
| Medical Long Context Reasoning (MLCR-AA, October 2026 methodology) | Health & medicine | 41.7 | 18.3 | −23.3 |
| GBENCH One-shot Coding GScore | Games | 78.4 | 56.4 | −22.0 |
| CritPt v2 (agentic, avg@5) | Science › Physical sciences | 83.1 | 62.3 | −20.9 |
| Artificial Analysis Intelligence Index v5.0.1 | Composite indexes | 56.9 | 36.0 | −20.8 |
| Terminal-Bench 4.0 | Software › Software engineering | 59.6 | 39.4 | −20.2 |
| EnterpriseBench: CoreCraft | Tool & computer use › Customer service & operations | 42.6 | 62.1 | +19.5 |
| BuseyBench SVG | Perception & media › Generation | 7.3 | 5.4 | −19.2 |
| AA-Omniscience v1.1 Index | Knowledge & reasoning › General knowledge | 44.6 | 7.7 | −18.4 |
| HLAB-Vals-1.1 | Professional › Legal | 7.1 | 25.0 | +17.9 |
| LMArena Agent Arena Praise vs Complaint | Tool & computer use › General agents | 32.3 | -2.7 | −17.5 |
| DrugDiscoveryBench | Science › Life sciences | 68.7 | 52.4 | −16.3 |
| FrontierCode 2.0 (Artificial Analysis) | Software › Software engineering | 42.4 | 27.2 | −15.2 |
| Harvey LAB (Mercor run, undisclosed ~213-task cohort) | Professional › Legal | 9.2 | 22.7 | +13.5 |
| SWE Atlas v1.1 - Overall | Software › Software engineering | 69.4 | 56.0 | −13.4 |
| DAYJOB: Finance (October 2026 rescoring) | Professional › Finance, accounting & tax | 37.3 | 24.2 | −13.1 |
| ComplexConstraints (November 2026 rescoring) | Language & writing › Instruction following | 70.8 | 58.8 | −12.0 |
| DAYJOB: Healthcare (October 2026 rescoring) | Health & medicine | 28.4 | 17.2 | −11.2 |
| SimpleBench | Knowledge & reasoning › General knowledge | 83.6 | 72.7 | −10.9 |
| Legal Research Bench | Professional › Legal | 39.4 | 49.0 | +9.6 |
| GDP.pdf (Artificial Analysis, All-pass) | Professional › General knowledge work | 32.2 | 22.8 | −9.4 |
| Chartography (November 2026 rescoring) | Perception & media › Documents & charts | 88.6 | 79.3 | −9.3 |
| HiL-Bench (Human-in-Loop Benchmark) | Tool & computer use › General agents | 65.3 | 74.0 | +8.7 |
| Professional Reasoning Benchmark - Finance | Professional › Finance, accounting & tax | 47.5 | 56.0 | +8.4 |
| LMArena Agent Arena Confirmed Success | Tool & computer use › General agents | 14.9 | -2.0 | −8.4 |
| Tax Agent Bench | Professional › Finance, accounting & tax | 63.3 | 71.4 | +8.0 |
| GDP.xlsx | Professional › General knowledge work | 22.9 | 15.4 | −7.5 |
| Artificial Analysis Progress Index v1 | Composite indexes | 350 | 329 | −7.3 |
| Vals Legal Index | Professional › Legal | 68.4 | 75.7 | +7.3 |
| Professional Reasoning Benchmark - Legal | Professional › Legal | 48.4 | 55.6 | +7.2 |
| τ³-Banking (Mercor run, bm25_grep retrieval) | Tool & computer use › Customer service & operations | 46.7 | 53.3 | +6.6 |
| OSWorld 2.1 full - reported partial reward, step limit unspecified | Tool & computer use › Computer & browser use | 75.8 | 69.5 | −6.3 |
| Vals Finance Index | Professional › Finance, accounting & tax | 55.6 | 60.9 | +5.3 |
| LMArena Agent Arena Steering Burden | Tool & computer use › General agents | 13.9 | 4.8 | −4.6 |
| APEX-SWE | Software › Software engineering | 48.6 | 52.9 | +4.3 |
| Finance Agent v2.1 | Professional › Finance, accounting & tax | 60.4 | 63.8 | +3.4 |
| AA-Briefcase v1.1 Elo | Professional › General knowledge work | 1569 | 1520 | −2.7 |
| HANDBOOK.md Agents v1.1 | Tool & computer use › Customer service & operations | 31.4 | 29.8 | −1.6 |
| GDP.pdf (November 2026 rescoring) | Professional › General knowledge work | 50.4 | 49.2 | −1.2 |
| APEX-1 (Mercor, mean score) | Professional › General knowledge work | 69.3 | 70.4 | +1.1 |
| APEX-Agents 1.1 | Professional › General knowledge work | 75.1 | 75.7 | +0.6 |
| Excel Modeling Benchmark v1.1 | Professional › Finance, accounting & tax | 72.7 | 72.2 | −0.5 |
| OSWorld 2.1 full - reported binary completion, step limit unspecified | Tool & computer use › Computer & browser use | 34.3 | 33.8 | −0.5 |
| AutomationBench-AA 1.0.6 | Tool & computer use › Customer service & operations | 68.5 | 68.1 | −0.4 |
| FORTRESS Average Risk Score (ARS) | Safety & alignment | 12.1 | 11.9 | +0.2 |
| Ramp SWE-Bench v2 (Bash harness) | Software › Software engineering | 54.6 | 54.4 | −0.2 |
| LMArena Agent Arena Tool Hallucination | Tool & computer use › Tool calling | 1.2 | 0.9 | −0.1 |
| LMArena Agent Arena Bash Reliability | Tool & computer use › General agents | 6.5 | 6.6 | +0.1 |
Data: Artificial Analysis, Surge AI, Vals AI and 12 othersatlas.village.dev