GPT-6.2 Sol vs GPT-6 Sol
GPT-6.2 Sol leads by 3.4 capability points (≈95% interval +0.9 to +5.9): a clear lead over GPT-6 Sol · P(higher) ≈ 100%
Wins 53 of 62 shared benchmarks (8 lost, 1 tied)
Capability vs. relative cost
GPT-6.2 SolGPT-6 SolClosed weightsOpen weights
Benchmark wins53 – 8 · 62 shared, 1 tied
Comparing matching setups only (1 benchmark excluded) · suites counted once
Area← GPT-6.2 SolGPT-6 Sol →
Data: Artificial Analysis, Vals AI, Arena and 15 othersatlas.village.dev
Margins
Each benchmark scaled 0–100% (its declared range, else the field’s range); Δ is GPT-6 Sol minus GPT-6.2 Sol.
| ARC-AGI-3 (Semi-Private) | Games | 98.2 | 23.0 | −75.2 |
| Terminal-Bench-Science 0.1 | Science › Multidisciplinary | 67.6 | 30.0 | −37.6 |
| AA-PersonalityBench Organization index | Language & writing › Style & personality | 45.7 | 12.4 | −33.6 |
| AA-PersonalityBench Imagination index | Language & writing › Style & personality | 60.1 | 29.6 | −30.8 |
| AA-PersonalityBench Conviction index | Language & writing › Style & personality | 79.2 | 50.5 | −29.0 |
| Terminal-Bench Science 0.1 (Vals) | Science › Multidisciplinary | 58.6 | 30.0 | −28.6 |
| AA-PersonalityBench Informativeness index | Language & writing › Style & personality | 37.7 | 10.5 | −27.5 |
| Humanity's Sixth Sense | Perception & media › Spatial & visual reasoning | 56.8 | 31.2 | −25.6 |
| Bug Hunt Bench | Software › Software engineering | 52.1 | 27.9 | −24.1 |
| AA-PersonalityBench Expressiveness index | Language & writing › Style & personality | 55.3 | 33.4 | −22.1 |
| Chartography | Perception & media › Documents & charts | 75.1 | 53.6 | −21.5 |
| Blueprint-Bench 2 | Perception & media › Spatial & visual reasoning | 0.6 | 0.4 | −20.1 |
| Humanity's Last Exam Diamond | Knowledge & reasoning › General knowledge | 64.0 | 45.2 | −18.8 |
| AA-PersonalityBench Transparency index | Language & writing › Style & personality | 56.9 | 39.1 | −18.0 |
| FrontierCUA v1 pass@1 | Tool & computer use › Computer & browser use | 25.0 | 7.5 | −17.5 |
| AA-3D v1.0 Elo | Software › Frontend & design | 1259 | 1131 | −17.4 |
| Terminal-Bench 4.0 | Software › Software engineering | 60.6 | 44.4 | −16.2 |
| Terminal-Bench 4.0 (Artificial Analysis) | Software › Software engineering | 60.1 | 43.9 | −16.2 |
| AA-PersonalityBench Slop index | Language & writing › Style & personality | 15.4 | 1.0 | −14.5 |
| DAYJOB: Finance (October 2026 rescoring) | Professional › Finance, accounting & tax | 32.8 | 19.5 | −13.3 |
| Artificial Analysis Intelligence Index v5.0.1 | Composite indexes | 57.3 | 44.2 | −13.1 |
| Vals Science Index | Science › Multidisciplinary | 57.8 | 45.0 | −12.8 |
| LMArena Code Arena WebDev | Software › Frontend & design | 1757 | 1674 | −11.0 |
| DAYJOB: Healthcare (October 2026 rescoring) | Health & medicine | 17.2 | 7.2 | −10.0 |
| Tax Agent Bench | Professional › Finance, accounting & tax | 62.9 | 53.0 | −9.9 |
| AutomationBench-AA 1.0.6 | Tool & computer use › Customer service & operations | 71.3 | 61.7 | −9.6 |
| AA-Omniscience v1.1 Index | Knowledge & reasoning › General knowledge | 40.4 | 21.4 | −9.5 |
| LMArena Agent Arena Net Improvement (Steering Burden) | Tool & computer use › General agents | 11.3 | 6.5 | −9.4 |
| FrontierCode 2.0 (Artificial Analysis) | Software › Software engineering | 44.4 | 35.6 | −8.8 |
| Legal Research Bench | Professional › Legal | 37.0 | 28.8 | −8.2 |
| AutomationBench 1.0.6 (Zapier) | Tool & computer use › Customer service & operations | 41.3 | 33.2 | −8.0 |
| DualEntry Accounting AI Benchmark | Professional › Finance, accounting & tax | 84.2 | 76.2 | −8.0 |
| Gray Swan IPI Attack Success (k=15) | Safety & alignment | 2.3 | 10.1 | −7.8 |
| Harvey LAB-AA v1.1 (hallucination-gated) | Professional › Legal | 11.1 | 3.6 | −7.5 |
| AA-Briefcase v1.1 Elo | Professional › General knowledge work | 1612 | 1476 | −7.5 |
| LMArena Agent Arena Praise vs Complaint | Tool & computer use › General agents | 23.3 | 9.8 | −6.8 |
| MysteryMechanism | Math | 36.9 | 30.2 | −6.8 |
| EnterpriseBench: CoreCraft | Tool & computer use › Customer service & operations | 36.4 | 43.1 | +6.7 |
| DeepsecBench — Revalidated V1 (233 issues) | Software › Cybersecurity | 47.3 | 40.7 | −6.6 |
| HealthBench Professional | Health & medicine | 65.5 | 59.5 | −6.0 |
| AA-PersonalityBench Warmth index | Language & writing › Style & personality | 60.1 | 54.3 | −5.8 |
| DeepsecBench-AA (Artificial Analysis) | Software › Cybersecurity | 51.6 | 46.3 | −5.3 |
| GDP.xlsx | Professional › General knowledge work | 22.3 | 17.1 | −5.2 |
| GDP.pdf (Artificial Analysis, All-pass) | Professional › General knowledge work | 30.0 | 25.2 | −4.8 |
| Professional Reasoning Benchmark - Legal | Professional › Legal | 43.3 | 38.6 | −4.7 |
| ARC-AGI-2 | Knowledge & reasoning › Abstract reasoning | 85.8 | 89.6 | +3.8 |
| LMArena Agent Arena Confirmed Success | Tool & computer use › General agents | 10.9 | 3.9 | −3.5 |
| Medical Long Context Reasoning (MLCR-AA, October 2026 methodology) | Health & medicine | 37.2 | 33.9 | −3.3 |
| MentalHealthBench | Health & medicine | 50.9 | 54.2 | +3.3 |
| SimpleBench | Knowledge & reasoning › General knowledge | 69.9 | 73.1 | +3.2 |
| BioMysteryBench (Vals Terminus 2) | Science › Life sciences | 77.8 | 74.8 | −3.0 |
| LMArena Agent Arena Bash Reliability | Tool & computer use › General agents | 4.3 | 10.0 | +2.8 |
| CWE-Bench-AA (Artificial Analysis) | Software › Cybersecurity | 61.7 | 64.2 | +2.5 |
| TxBench Preclinical Pharmacology V1 Clean Reruns | Science › Life sciences | 65.3 | 63.0 | −2.3 |
| CritPt v2 (agentic, avg@5) | Science › Physical sciences | 78.6 | 80.9 | +2.3 |
| LMArena Agent Arena Steering Burden | Tool & computer use › General agents | 13.4 | 9.3 | −2.1 |
| ARC-AGI-1 Semi-Private | Knowledge & reasoning › Abstract reasoning | 97.5 | 95.5 | −2.0 |
| TaxCalcBench TY25 | Professional › Finance, accounting & tax | 32.0 | 30.0 | −2.0 |
| Artificial Analysis Cyber Index v1 | Software › Cybersecurity | 37.7 | 36.8 | −0.9 |
| Vibe Code Bench v1.1 | Software › Frontend & design | 87.6 | 87.8 | +0.2 |
| LMArena Agent Arena Tool Hallucination | Tool & computer use › Tool calling | 1.2 | 1.2 | 0.0 |
| CyberGym-E2E-AA (Artificial Analysis) | Software › Cybersecurity | 0.0 | 0.0 | 0.0 |
Data: Artificial Analysis, Vals AI, Arena and 15 othersatlas.village.dev