GPT-6.2 Astra vs Grok 4.6
GPT-6.2 Astra leads by 8.8 capability points (≈95% interval +6.4 to +11.2): a clear lead over Grok 4.6 · P(higher) ≈ 100%
Wins 51 of 61 shared benchmarks (9 lost, 1 tied)
Capability vs. relative cost
GPT-6.2 AstraGrok 4.6Closed weightsOpen weights
Benchmark wins51 – 9 · 61 shared, 1 tied
Comparing matching setups only (10 benchmarks excluded) · suites counted once
Area← GPT-6.2 AstraGrok 4.6 →
Data: Artificial Analysis, Epoch AI, Vals AI and 13 othersatlas.village.dev
Margins
Each benchmark scaled 0–100% (its declared range, else the field’s range); Δ is Grok 4.6 minus GPT-6.2 Astra.
| Terminal-Bench-Science 0.1 | Science › Multidisciplinary | 78.6 | 2.9 | −75.7 |
| FrontierMath Tier 4 (v2) - Private | Math | 100.0 | 31.7 | −68.3 |
| Mystery Game Puzzles | Games | 95.0 | 34.0 | −61.0 |
| Terminal-Bench Science 0.1 (Vals) | Science › Multidisciplinary | 71.4 | 11.4 | −60.0 |
| ARC-AGI-3 (Semi-Private) | Games | 56.5 | 2.1 | −54.4 |
| Humanity's Last Exam Diamond | Knowledge & reasoning › General knowledge | 72.2 | 21.5 | −50.7 |
| Gray Swan IPI Attack Success (k=15) | Safety & alignment | 2.4 | 51.8 | −49.4 |
| Furniture Assembly Benchmark | Perception & media › Spatial & visual reasoning | 89.2 | 40.0 | −49.2 |
| Terminal-Bench 4.0 (Artificial Analysis) | Software › Software engineering | 64.1 | 17.2 | −47.0 |
| Chartography (November 2026 rescoring) | Perception & media › Documents & charts | 90.7 | 44.4 | −46.3 |
| Terminal-Bench 4.0 | Software › Software engineering | 63.1 | 17.2 | −46.0 |
| DeepsecBench — Revalidated V1 (233 issues) | Software › Cybersecurity | 61.2 | 16.1 | −45.1 |
| Chess Puzzles (task version 2.0) | Games | 78.0 | 33.0 | −45.0 |
| AA-PersonalityBench Imagination index | Language & writing › Style & personality | 63.3 | 20.0 | −43.7 |
| AA-PersonalityBench Warmth index | Language & writing › Style & personality | 68.1 | 31.5 | −37.0 |
| Artificial Analysis Intelligence Index v5.0.1 | Composite indexes | 62.1 | 31.0 | −31.1 |
| AA-PersonalityBench Transparency index | Language & writing › Style & personality | 40.9 | 71.4 | +30.8 |
| LMArena Agent Arena Net Improvement (Steering Burden) | Tool & computer use › General agents | 15.2 | -0.6 | −30.8 |
| Vals Science Index | Science › Multidisciplinary | 68.4 | 38.1 | −30.3 |
| SimpleQA Verified (Epoch anti-abstention) | Knowledge & reasoning › General knowledge | 78.5 | 49.3 | −29.2 |
| FrontierCode 2.0 (Artificial Analysis) | Software › Software engineering | 49.0 | 19.8 | −29.2 |
| ARC-AGI-2 | Knowledge & reasoning › Abstract reasoning | 95.8 | 67.1 | −28.7 |
| Blueprint-Bench 2 | Perception & media › Spatial & visual reasoning | 0.6 | 0.3 | −28.2 |
| AA-PersonalityBench Slop index | Language & writing › Style & personality | 7.4 | 35.3 | +28.2 |
| MirrorCode (ML, +Private, 2L) | Software › Software engineering | 67.8 | 41.1 | −26.7 |
| FrontierMath Tiers 1-3 (v2) - Private | Math | 92.6 | 66.0 | −26.7 |
| DAYJOB: Finance (October 2026 rescoring) | Professional › Finance, accounting & tax | 45.8 | 21.0 | −24.8 |
| EnterpriseOps-Gym-AA | Tool & computer use › Customer service & operations | 25.2 | 48.3 | +23.2 |
| MysteryMechanism | Math | 53.2 | 30.6 | −22.5 |
| EnterpriseBench: CoreCraft | Tool & computer use › Customer service & operations | 43.6 | 65.6 | +22.0 |
| LiveBench 2.0 (2026-11-06 release) | Composite indexes | 84.9 | 63.5 | −21.4 |
| AA-PersonalityBench Conviction index | Language & writing › Style & personality | 71.3 | 92.4 | +21.3 |
| Medical Long Context Reasoning (MLCR-AA, October 2026 methodology) | Health & medicine | 39.4 | 18.3 | −21.1 |
| CritPt v2 (agentic, avg@5) | Science › Physical sciences | 83.1 | 62.3 | −20.9 |
| LMArena Agent Arena Praise vs Complaint | Tool & computer use › General agents | 37.0 | -3.5 | −20.3 |
| AA-AnalystAgent (pass^5) | Professional › General knowledge work | 61.3 | 41.3 | −20.0 |
| DAYJOB: Healthcare (October 2026 rescoring) | Health & medicine | 34.4 | 14.8 | −19.6 |
| HealthBench Professional (Length-Adjusted) | Health & medicine | 66.3 | 51.7 | −14.6 |
| Vibe Code Bench v1.1 | Software › Frontend & design | 89.9 | 76.2 | −13.7 |
| AA-PersonalityBench Informativeness index | Language & writing › Style & personality | 31.3 | 44.8 | +13.6 |
| Tax Agent Bench | Professional › Finance, accounting & tax | 57.9 | 71.4 | +13.5 |
| LLM Debate Benchmark | Language & writing › Language & wordplay | 1673 | 1585 | −12.6 |
| AA-Omniscience v1.1 Index | Knowledge & reasoning › General knowledge | 50.0 | 24.9 | −12.6 |
| Epoch Capabilities Index (ECI) | Composite indexes | 173 | 158 | −12.0 |
| GDP.pdf (Artificial Analysis, All-pass) | Professional › General knowledge work | 29.8 | 17.8 | −12.0 |
| ARC-AGI-1 Semi-Private | Knowledge & reasoning › Abstract reasoning | 98.5 | 87.5 | −11.0 |
| AutomationBench-AA 1.0.6 | Tool & computer use › Customer service & operations | 77.2 | 67.0 | −10.3 |
| MMMU Pro | Knowledge & reasoning › General knowledge | 89.5 | 80.5 | −9.0 |
| BioMysteryBench (Vals Terminus 2) | Science › Life sciences | 80.7 | 72.2 | −8.5 |
| LMArena Agent Arena Steering Burden | Tool & computer use › General agents | 14.0 | -2.5 | −8.2 |
| AA-PersonalityBench Organization index | Language & writing › Style & personality | 21.8 | 14.3 | −7.5 |
| LMArena Agent Arena Confirmed Success | Tool & computer use › General agents | 12.4 | -2.5 | −7.5 |
| TxBench Preclinical Pharmacology V1 Clean Reruns | Science › Life sciences | 68.0 | 61.7 | −6.3 |
| Legal Research Bench | Professional › Legal | 42.8 | 48.1 | +5.3 |
| AA-Briefcase v1.1 Elo | Professional › General knowledge work | 1594 | 1540 | −3.0 |
| GPQA Diamond | Science › Multidisciplinary | 96.8 | 94.0 | −2.8 |
| AA-PersonalityBench Expressiveness index | Language & writing › Style & personality | 66.5 | 63.8 | −2.7 |
| SimpleBench | Knowledge & reasoning › General knowledge | 77.3 | 75.9 | −1.4 |
| OTIS Mock AIME 2024-2025 | Math | 100.0 | 99.2 | −0.8 |
| LMArena Agent Arena Bash Reliability | Tool & computer use › General agents | 7.7 | 9.1 | +0.7 |
| LMArena Agent Arena Tool Hallucination | Tool & computer use › Tool calling | 1.2 | 1.2 | 0.0 |
Data: Artificial Analysis, Epoch AI, Vals AI and 13 othersatlas.village.dev