Gemini 4 Carbon vs GPT-6.2 Astra
GPT-6.2 Astra leads by 0.8 capability points (≈95% interval −2.0 to +3.7): statistically tied · P(higher) ≈ 72%
Wins 44 of 84 shared benchmarks (36 lost, 4 tied)
Capability vs. relative cost
Gemini 4 CarbonGPT-6.2 AstraClosed weightsOpen weights
Benchmark wins36 – 44 · 84 shared, 4 tied
Comparing matching setups only (41 benchmarks excluded) · suites counted once
Area← Gemini 4 CarbonGPT-6.2 Astra →
Data: Artificial Analysis, Surge AI, Epoch AI and 18 othersatlas.village.dev
Margins
Each benchmark scaled 0–100% (its declared range, else the field’s range); Δ is GPT-6.2 Astra minus Gemini 4 Carbon.
| KradleBench | Games | 22.0 | 78.3 | +56.3 |
| EnterpriseBench: CoreCraft | Tool & computer use › Customer service & operations | 80.5 | 43.6 | −36.9 |
| DAYJOB: Law (November 2026 release) | Professional › Legal | 13.8 | 41.4 | +27.6 |
| DeepsecBench — Revalidated V1 (233 issues) | Software › Cybersecurity | 35.2 | 61.2 | +26.0 |
| Mystery Game Puzzles | Games | 73.0 | 95.0 | +22.0 |
| DAYJOB: Finance (October 2026 rescoring) | Professional › Finance, accounting & tax | 23.8 | 45.8 | +22.0 |
| Medical Long Context Reasoning (MLCR-AA, October 2026 methodology) | Health & medicine | 19.4 | 39.4 | +20.0 |
| Tax Agent Bench | Professional › Finance, accounting & tax | 76.4 | 57.9 | −18.6 |
| DAYJOB: Accounting | Professional › Finance, accounting & tax | 25.2 | 42.0 | +16.8 |
| SimpleBench | Knowledge & reasoning › General knowledge | 92.6 | 77.3 | −15.3 |
| Legal Research Bench | Professional › Legal | 55.8 | 42.8 | −13.0 |
| MultiChallenge | Language & writing › Conversation | 85.9 | 73.2 | −12.7 |
| Terminal-Bench-Science 0.1 | Science › Multidisciplinary | 62.4 | 74.8 | +12.4 |
| Furniture Assembly Benchmark | Perception & media › Spatial & visual reasoning | 77.5 | 89.2 | +11.7 |
| Dual-Arm Sim (task completion) | Tool & computer use › Robotics | 13.8 | 2.5 | −11.3 |
| SWE Atlas v1.1 - Overall | Software › Software engineering | 71.5 | 60.5 | −11.0 |
| Professional Reasoning Benchmark - Legal | Professional › Legal | 53.4 | 43.1 | −10.3 |
| Vals Finance Index | Professional › Finance, accounting & tax | 67.0 | 57.3 | −9.7 |
| GDP.xlsx | Professional › General knowledge work | 38.9 | 29.7 | −9.2 |
| Humanity's Sixth Sense | Perception & media › Spatial & visual reasoning | 71.1 | 63.2 | −7.9 |
| sudo L7 (October 2026 rescoring) | Software › Software engineering | 30.6 | 38.3 | +7.7 |
| FrontierMath Erdős | Math | 2.9 | 10.3 | +7.4 |
| GDP.pdf (November 2026 rescoring) | Professional › General knowledge work | 56.4 | 49.2 | −7.2 |
| LMArena Agent Arena Bash Reliability | Tool & computer use › General agents | -5.7 | 7.7 | +6.7 |
| FrontierCode 2.0 (Artificial Analysis) | Software › Software engineering | 40.6 | 47.1 | +6.6 |
| Finance Agent v2.1 | Professional › Finance, accounting & tax | 67.2 | 60.7 | −6.4 |
| Vals Legal Index | Professional › Legal | 75.9 | 69.7 | −6.2 |
| DrugDiscoveryBench | Science › Life sciences | 71.5 | 65.5 | −6.0 |
| Chartography (November 2026 rescoring) | Perception & media › Documents & charts | 85.5 | 90.7 | +5.2 |
| FrontierMath Tier 4 (v2) - Private | Math | 95.1 | 100.0 | +4.9 |
| EBR-bench v4 (card ban, single agent) | Games | 87.6 | 92.4 | +4.8 |
| DAYJOB: Tax | Professional › Finance, accounting & tax | 4.8 | 9.2 | +4.4 |
| HANDBOOK.md Agents v1.1 | Tool & computer use › Customer service & operations | 49.8 | 45.8 | −4.0 |
| RiemannBench (November 2026 rescoring) | Math | 80.0 | 76.0 | −4.0 |
| BigFinanceBench private holdout (final-answer accuracy) | Professional › Finance, accounting & tax | 50.1 | 54.0 | +3.9 |
| VIALS-AA (pass@3) | Science › Life sciences | 70.8 | 74.5 | +3.7 |
| Code Migration v1.1 (Vals Index v2.2 subset) | Software › Software engineering | 80.1 | 83.6 | +3.5 |
| Artificial Analysis Legal Index | Professional › Legal | 62.0 | 58.7 | −3.2 |
| Humanity's Last Exam Diamond | Knowledge & reasoning › General knowledge | 69.1 | 72.2 | +3.1 |
| VIALS-AA (average of 3) | Science › Life sciences | 67.3 | 70.2 | +2.9 |
| Blueprint-Bench 2 | Perception & media › Spatial & visual reasoning | 0.6 | 0.6 | +2.8 |
| Epoch Capabilities Index (ECI) | Composite indexes | 170 | 173 | +2.5 |
| LMArena Agent Arena Confirmed Success | Tool & computer use › General agents | 17.2 | 12.4 | −2.4 |
| Artificial Analysis Strategy & Ops Index | Professional › General knowledge work | 60.9 | 63.3 | +2.3 |
| AA-Briefcase v1.1 Elo | Professional › General knowledge work | 1611 | 1570 | −2.2 |
| Artificial Analysis Intelligence Index v5.0.1 | Composite indexes | 60.0 | 62.1 | +2.1 |
| FrontierMath Tiers 1-3 (v2) - Private | Math | 90.5 | 92.6 | +2.1 |
| Terminal-Bench 4.0 (Artificial Analysis) | Software › Software engineering | 60.6 | 58.6 | −2.0 |
| TxBench Preclinical Pharmacology V1 Clean Reruns | Science › Life sciences | 66.0 | 68.0 | +2.0 |
| LMArena Agent Arena Steering Burden | Tool & computer use › General agents | 17.9 | 14.0 | −2.0 |
| VIALS-AA (pass^3) | Science › Life sciences | 64.0 | 65.8 | +1.9 |
| AA-Omniscience v1.1 Index | Knowledge & reasoning › General knowledge | 53.5 | 50.0 | −1.7 |
| LMArena Agent Arena Net Improvement (Steering Burden) | Tool & computer use › General agents | 16.1 | 15.2 | −1.7 |
| Artificial Analysis Finance & Accounting Index | Professional › Finance, accounting & tax | 62.6 | 60.9 | −1.7 |
| MKA (Artificial Analysis) | Professional › General knowledge work | 54.3 | 52.7 | −1.7 |
| ARC-AGI-2 | Knowledge & reasoning › Abstract reasoning | 94.2 | 95.8 | +1.7 |
| GDPval v2 (OpenAI internal) | Professional › General knowledge work | 48.8 | 50.4 | +1.6 |
| LiveBench 2.0 (2026-11-06 release) | Composite indexes | 83.3 | 84.9 | +1.5 |
| Artificial Analysis Economics Index | Composite indexes | 59.6 | 61.1 | +1.5 |
| SEC-Bench Pro - release-reported cohort | Software › Cybersecurity | 88.2 | 86.8 | −1.5 |
| CritPt v2 (agentic, avg@5) | Science › Physical sciences | 82.3 | 83.7 | +1.4 |
| Artificial Analysis Engineering Index | Science › Engineering | 65.3 | 66.6 | +1.3 |
| ComplexConstraints (November 2026 rescoring) | Language & writing › Instruction following | 70.9 | 69.6 | −1.3 |
| ARC-AGI-3 (Semi-Private) | Games | 99.1 | 97.8 | −1.3 |
| HLAB-Vals-1.1 | Professional › Legal | 17.5 | 16.3 | −1.2 |
| LMArena Agent Arena Praise vs Complaint | Tool & computer use › General agents | 39.4 | 37.0 | −1.2 |
| Ramp SWE-Bench v2 (Bash harness) | Software › Software engineering | 74.0 | 75.0 | +1.0 |
| Agents' Last Exam v1.1 Score | Professional › General knowledge work | 77.4 | 78.3 | +0.9 |
| Vibe Code Bench v1.1 | Software › Frontend & design | 89.1 | 89.9 | +0.8 |
| DAYJOB: Healthcare (October 2026 rescoring) | Health & medicine | 35.2 | 34.4 | −0.8 |
| SimpleQA Verified (Epoch anti-abstention) | Knowledge & reasoning › General knowledge | 77.7 | 78.5 | +0.8 |
| AutomationBench 1.0.6 (Zapier) | Tool & computer use › Customer service & operations | 53.0 | 52.4 | −0.6 |
| AutomationBench-AA 1.0.6 | Tool & computer use › Customer service & operations | 77.8 | 77.2 | −0.6 |
| Excel Modeling Benchmark v1.1 | Professional › Finance, accounting & tax | 76.5 | 76.0 | −0.4 |
| LLM Debate Benchmark | Language & writing › Language & wordplay | 1676 | 1673 | −0.4 |
| GPQA Diamond | Science › Multidisciplinary | 96.5 | 96.8 | +0.3 |
| MMMU Pro | Knowledge & reasoning › General knowledge | 89.2 | 89.5 | +0.2 |
| Artificial Analysis Healthcare & Medical Index | Health & medicine | 56.8 | 57.0 | +0.2 |
| GDP.pdf (Artificial Analysis, All-pass) | Professional › General knowledge work | 29.6 | 29.8 | +0.2 |
| LMArena Agent Arena Tool Hallucination | Tool & computer use › Tool calling | 0.8 | 1.2 | +0.2 |
| ARC-AGI-1 Semi-Private | Knowledge & reasoning › Abstract reasoning | 98.5 | 98.5 | 0.0 |
| Chess Puzzles (task version 2.0) | Games | 78.0 | 78.0 | 0.0 |
| OTIS Mock AIME 2024-2025 | Math | 100.0 | 100.0 | 0.0 |
| Terminal-Bench 4.0 | Software › Software engineering | 63.1 | 63.1 | 0.0 |
Data: Artificial Analysis, Surge AI, Epoch AI and 18 othersatlas.village.dev