GPT-6.2 Astra vs Muse Spark 1.3
GPT-6.2 Astra leads by 7.5 capability points (≈95% interval +5.0 to +10.0): a clear lead over Muse Spark 1.3 · P(higher) ≈ 100%
Wins 51 of 67 shared benchmarks (16 lost)
Capability vs. relative cost
GPT-6.2 AstraMuse Spark 1.3Closed weightsOpen weights
Benchmark wins51 – 16 · 67 shared
Comparing matching setups only (16 benchmarks excluded) · suites counted once
Area← GPT-6.2 AstraMuse Spark 1.3 →
Data: Artificial Analysis, Surge AI, Epoch AI and 9 othersatlas.village.dev
Margins
Each benchmark scaled 0–100% (its declared range, else the field’s range); Δ is Muse Spark 1.3 minus GPT-6.2 Astra.
| Mystery Game Puzzles | Games | 95.0 | 25.0 | −70.0 |
| AA-PersonalityBench Informativeness index | Language & writing › Style & personality | 31.3 | 100 | +69.4 |
| AA-PersonalityBench Slop index | Language & writing › Style & personality | 7.4 | 75.3 | +68.5 |
| Terminal-Bench-Science 0.1 | Science › Multidisciplinary | 78.1 | 11.0 | −67.1 |
| Terminal-Bench Science 0.1 (Vals) | Science › Multidisciplinary | 71.4 | 10.0 | −61.4 |
| CyberGym-E2E-AA (Artificial Analysis) | Software › Cybersecurity | 0.0 | 59.5 | +59.5 |
| FrontierMath Tier 4 (v2) - Private | Math | 100.0 | 46.3 | −53.7 |
| AA-PersonalityBench Organization index | Language & writing › Style & personality | 21.8 | 63.8 | +42.5 |
| Chess Puzzles (task version 2.0) | Games | 78.0 | 37.0 | −41.0 |
| DeepsecBench-AA (Artificial Analysis) | Software › Cybersecurity | 52.7 | 12.5 | −40.2 |
| AA-3D v1.0 Elo | Software › Frontend & design | 1318 | 1032 | −38.9 |
| Humanity's Last Exam Diamond | Knowledge & reasoning › General knowledge | 72.2 | 33.7 | −38.5 |
| Terminal-Bench 4.0 | Software › Software engineering | 63.1 | 24.7 | −38.4 |
| TaxCalcBench TY25 | Professional › Finance, accounting & tax | 44.0 | 6.0 | −38.0 |
| AA-PersonalityBench Imagination index | Language & writing › Style & personality | 63.3 | 98.1 | +35.2 |
| AutomationBench 1.0.6 (Zapier) | Tool & computer use › Customer service & operations | 52.4 | 20.7 | −31.7 |
| Terminal-Bench 4.0 (Artificial Analysis) | Software › Software engineering | 64.1 | 33.3 | −30.8 |
| DAYJOB: Law (November 2026 release) | Professional › Legal | 41.4 | 11.8 | −29.6 |
| AA-PersonalityBench Warmth index | Language & writing › Style & personality | 68.1 | 96.2 | +28.4 |
| LMArena Agent Arena Net Improvement (Steering Burden) | Tool & computer use › General agents | 15.2 | 1.1 | −27.5 |
| Humanity's Sixth Sense | Perception & media › Spatial & visual reasoning | 63.2 | 37.4 | −25.8 |
| Artificial Analysis Intelligence Index v5.0.1 | Composite indexes | 62.1 | 36.6 | −25.5 |
| sudo L7 (October 2026 rescoring) | Software › Software engineering | 38.3 | 12.8 | −25.5 |
| AA-AnalystAgent (pass^5) | Professional › General knowledge work | 62.5 | 37.5 | −25.0 |
| DAYJOB: Accounting | Professional › Finance, accounting & tax | 42.0 | 18.0 | −24.0 |
| Medical Long Context Reasoning (MLCR-AA, October 2026 methodology) | Health & medicine | 39.4 | 15.6 | −23.9 |
| RiemannBench (November 2026 rescoring) | Math | 76.0 | 53.6 | −22.4 |
| FrontierCode 2.0 (Artificial Analysis) | Software › Software engineering | 45.4 | 23.1 | −22.3 |
| AutomationBench-AA 1.0.6 | Tool & computer use › Customer service & operations | 77.2 | 57.9 | −19.4 |
| DAYJOB: Finance (October 2026 rescoring) | Professional › Finance, accounting & tax | 45.8 | 27.0 | −18.8 |
| LMArena Agent Arena Praise vs Complaint | Tool & computer use › General agents | 37.0 | 0.1 | −18.5 |
| Professional Reasoning Benchmark - Legal | Professional › Legal | 43.1 | 61.6 | +18.5 |
| EnterpriseBench: CoreCraft | Tool & computer use › Customer service & operations | 43.6 | 61.9 | +18.3 |
| FrontierMath Tiers 1-3 (v2) - Private | Math | 92.6 | 74.4 | −18.2 |
| MysteryMechanism | Math | 53.2 | 36.0 | −17.1 |
| LiveBench 2.0 (2026-11-06 release) | Composite indexes | 84.9 | 68.6 | −16.3 |
| CritPt v2 (agentic, avg@5) | Science › Physical sciences | 83.7 | 68.0 | −15.7 |
| Tax Agent Bench | Professional › Finance, accounting & tax | 57.9 | 72.4 | +14.6 |
| LLM Debate Benchmark | Language & writing › Language & wordplay | 1673 | 1578 | −13.7 |
| Gray Swan IPI Attack Success (k=15) | Safety & alignment | 2.4 | 15.9 | −13.5 |
| MirrorCode (ML, +Private, 2L) | Software › Software engineering | 67.8 | 54.4 | −13.3 |
| AA-Omniscience v1.1 Index | Knowledge & reasoning › General knowledge | 50.0 | 23.8 | −13.1 |
| Legal Research Bench | Professional › Legal | 42.8 | 55.3 | +12.5 |
| DAYJOB: Healthcare (October 2026 rescoring) | Health & medicine | 34.4 | 22.0 | −12.4 |
| HANDBOOK.md Agents v1.1 | Tool & computer use › Customer service & operations | 45.8 | 33.9 | −11.9 |
| Epoch Capabilities Index (ECI) | Composite indexes | 173 | 158 | −11.7 |
| CWE-Bench-AA (Artificial Analysis) | Software › Cybersecurity | 69.2 | 60.8 | −8.3 |
| GDP.xlsx | Professional › General knowledge work | 29.7 | 22.6 | −7.1 |
| LMArena Agent Arena Steering Burden | Tool & computer use › General agents | 14.0 | 0.8 | −6.6 |
| Harvey LAB-AA v1.1 (hallucination-gated) | Professional › Legal | 15.0 | 8.9 | −6.1 |
| Chartography (November 2026 rescoring) | Perception & media › Documents & charts | 90.7 | 84.6 | −6.1 |
| AA-PersonalityBench Conviction index | Language & writing › Style & personality | 71.3 | 77.2 | +6.0 |
| DAYJOB: Tax | Professional › Finance, accounting & tax | 9.2 | 3.6 | −5.6 |
| AA-PersonalityBench Expressiveness index | Language & writing › Style & personality | 66.5 | 71.4 | +5.0 |
| LMArena Agent Arena Confirmed Success | Tool & computer use › General agents | 12.4 | 2.5 | −5.0 |
| ComplexConstraints (November 2026 rescoring) | Language & writing › Instruction following | 69.6 | 64.8 | −4.8 |
| SimpleBench | Knowledge & reasoning › General knowledge | 77.3 | 81.8 | +4.5 |
| Vibe Code Bench v1.1 | Software › Frontend & design | 89.9 | 85.9 | −4.0 |
| GDP.pdf (November 2026 rescoring) | Professional › General knowledge work | 49.2 | 53.0 | +3.8 |
| Artificial Analysis Cyber Index v1 | Software › Cybersecurity | 40.6 | 44.3 | +3.7 |
| AA-Briefcase v1.1 Elo | Professional › General knowledge work | 1644 | 1581 | −3.4 |
| LMArena Agent Arena Bash Reliability | Tool & computer use › General agents | 7.7 | 1.2 | −3.3 |
| DrugDiscoveryBench | Science › Life sciences | 65.5 | 62.2 | −3.2 |
| GDP.pdf (Artificial Analysis, All-pass) | Professional › General knowledge work | 29.8 | 26.6 | −3.2 |
| OTIS Mock AIME 2024-2025 | Math | 100.0 | 99.2 | −0.8 |
| LMArena Agent Arena Tool Hallucination | Tool & computer use › Tool calling | 1.2 | 1.0 | −0.1 |
| AA-PersonalityBench Transparency index | Language & writing › Style & personality | 40.9 | 41.0 | +0.1 |
Data: Artificial Analysis, Surge AI, Epoch AI and 9 othersatlas.village.dev