Gemini 4 Hydrogen vs GPT-6 Astra
TryTop 2 now
GPT-6 Astra leads by 1.4 capability points (≈95% interval −1.2 to +4.1): statistically tied · P(higher) ≈ 85%
Wins 38 of 76 shared benchmarks (36 lost, 2 tied)
Capability vs. relative cost
Gemini 4 HydrogenGPT-6 AstraClosed weightsOpen weights
Benchmark wins36 – 38 · 76 shared, 2 tied
Comparing matching setups only (35 benchmarks excluded) · suites counted once
Area← Gemini 4 HydrogenGPT-6 Astra →
Data: Artificial Analysis, Epoch AI, Surge AI and 19 othersatlas.village.dev
Margins
Each benchmark scaled 0–100% (its declared range, else the field’s range); Δ is GPT-6 Astra minus Gemini 4 Hydrogen.
| TxBench-Antibody-Development V1 (60 evaluations) | Science › Life sciences | 60.6 | 32.8 | −27.8 |
| DAYJOB: Law (November 2026 release) | Professional › Legal | 9.7 | 36.6 | +26.9 |
| SpireClimb | Games | 7.7 | 34.0 | +26.3 |
| Furniture Assembly Benchmark | Perception & media › Spatial & visual reasoning | 57.5 | 80.0 | +22.5 |
| Mystery Game Puzzles | Games | 65.0 | 84.0 | +19.0 |
| GDP.xlsx | Professional › General knowledge work | 40.9 | 22.9 | −18.0 |
| Ramp SWE-Bench v2 (Bash harness) | Software › Software engineering | 70.8 | 54.6 | −16.2 |
| Artificial Analysis Intelligence Index v5.0.1 | Composite indexes | 41.2 | 56.9 | +15.6 |
| Tax Agent Bench | Professional › Finance, accounting & tax | 78.4 | 63.3 | −15.1 |
| Medical Long Context Reasoning (MLCR-AA, October 2026 methodology) | Health & medicine | 27.2 | 41.7 | +14.4 |
| Legal Research Bench | Professional › Legal | 53.8 | 39.4 | −14.4 |
| Chess Puzzles (task version 2.0) | Games | 60.0 | 74.0 | +14.0 |
| HANDBOOK.md Agents v1.1 | Tool & computer use › Customer service & operations | 44.6 | 31.4 | −13.2 |
| FrontierSWE V2 Best@5 | Software › Software engineering | 61.2 | 73.0 | +11.8 |
| GBENCH One-shot Coding GScore | Games | 66.8 | 78.4 | +11.5 |
| Vals Finance Index | Professional › Finance, accounting & tax | 66.1 | 55.6 | −10.5 |
| HiL-Bench (Human-in-Loop Benchmark) | Tool & computer use › General agents | 55.3 | 65.3 | +10.0 |
| FrontierMath Tier 4 (v2) - Private | Math | 87.8 | 97.6 | +9.8 |
| FrontierSWE V2 Worst@5 | Software › Software engineering | 45.3 | 55.1 | +9.8 |
| Humanity's Sixth Sense | Perception & media › Spatial & visual reasoning | 63.4 | 53.6 | −9.8 |
| AutomationBench 1.0.6 (Zapier) | Tool & computer use › Customer service & operations | 51.1 | 41.4 | −9.7 |
| Gray Swan IPI Attack Success (k=15) | Safety & alignment | 0.0 | 8.5 | −8.5 |
| SimpleBench | Knowledge & reasoning › General knowledge | 91.7 | 83.6 | −8.1 |
| LiveBench 2.0 (2026-11-06 release) | Composite indexes | 77.0 | 84.3 | +7.3 |
| Terminal-Bench 4.0 (Artificial Analysis) | Software › Software engineering | 53.5 | 59.6 | +6.1 |
| AutomationBench-AA 1.0.6 | Tool & computer use › Customer service & operations | 74.4 | 68.5 | −5.9 |
| EBR-bench v4 (card ban, single agent) | Games | 70.5 | 76.2 | +5.7 |
| FrontierMath Tiers 1-3 (v2) - Private | Math | 88.1 | 93.7 | +5.6 |
| Terminal-Bench 4.0 | Software › Software engineering | 54.0 | 59.6 | +5.6 |
| LMArena Agent Arena Bash Reliability | Tool & computer use › General agents | -4.5 | 6.5 | +5.5 |
| FORTRESS Average Risk Score (ARS) | Safety & alignment | 17.5 | 12.1 | +5.3 |
| Vals Legal Index | Professional › Legal | 73.6 | 68.4 | −5.2 |
| Terminal-Bench-Science 0.1 | Science › Multidisciplinary | 49.5 | 54.8 | +5.2 |
| BigFinanceBench private holdout (final-answer accuracy) | Professional › Finance, accounting & tax | 49.4 | 54.6 | +5.2 |
| Humanity's Last Exam Diamond | Knowledge & reasoning › General knowledge | 58.8 | 63.8 | +5.0 |
| Professional Reasoning Benchmark - Finance | Professional › Finance, accounting & tax | 52.3 | 47.5 | −4.8 |
| Professional Reasoning Benchmark - Legal | Professional › Legal | 53.1 | 48.4 | −4.7 |
| micro1 PharmaBench | Professional › General knowledge work | 11.4 | 16.1 | +4.7 |
| SimpleQA Verified (Epoch anti-abstention) | Knowledge & reasoning › General knowledge | 70.9 | 75.6 | +4.7 |
| Finance Agent v2.1 | Professional › Finance, accounting & tax | 65.1 | 60.4 | −4.7 |
| Code Migration v1.1 (Vals Index v2.2 subset) | Software › Software engineering | 82.7 | 78.0 | −4.6 |
| BuseyBench SVG | Perception & media › Generation | 6.9 | 7.3 | +4.6 |
| Chartography | Perception & media › Documents & charts | 66.6 | 71.0 | +4.4 |
| DAYJOB: Healthcare (October 2026 rescoring) | Health & medicine | 24.0 | 28.4 | +4.4 |
| HLAB-Vals-1.1 | Professional › Legal | 11.3 | 7.1 | −4.2 |
| GDP.pdf (November 2026 rescoring) | Professional › General knowledge work | 54.2 | 50.4 | −3.8 |
| DAYJOB: Accounting | Professional › Finance, accounting & tax | 33.6 | 30.0 | −3.6 |
| Excel Modeling Benchmark v1.1 | Professional › Finance, accounting & tax | 75.8 | 72.7 | −3.1 |
| ComplexConstraints (November 2026 rescoring) | Language & writing › Instruction following | 73.7 | 70.8 | −2.9 |
| CritPt v2 (agentic, avg@5) | Science › Physical sciences | 80.6 | 83.4 | +2.9 |
| LMArena Text Arena Elo | Language & writing › General assistance | 1490 | 1475 | −2.6 |
| ARC-AGI-2 | Knowledge & reasoning › Abstract reasoning | 92.5 | 95.0 | +2.5 |
| Dual-Arm Sim (task completion) | Tool & computer use › Robotics | 0.0 | 2.5 | +2.5 |
| FrontierCode 2.0 (Artificial Analysis) | Software › Software engineering | 39.0 | 41.4 | +2.4 |
| TxBench Oligo Discovery V1 (113 evaluations) | Science › Life sciences | 54.3 | 51.9 | −2.4 |
| LMArena Agent Arena Net Improvement (Steering Burden) | Tool & computer use › General agents | 13.5 | 14.6 | +2.0 |
| TxBench Preclinical Pharmacology V1 Clean Reruns | Science › Life sciences | 61.3 | 59.3 | −2.0 |
| DualEntry Accounting AI Benchmark | Professional › Finance, accounting & tax | 83.2 | 81.2 | −2.0 |
| TaxCalcBench TY25 | Professional › Finance, accounting & tax | 44.0 | 46.0 | +2.0 |
| Epoch Capabilities Index (ECI) | Composite indexes | 167 | 169 | +1.9 |
| LMArena Agent Arena Steering Burden | Tool & computer use › General agents | 17.4 | 13.9 | −1.7 |
| GradeBench | Professional › Education | 51.9 | 53.6 | +1.7 |
| TxBench Antibody Discovery V1 (100 evaluations) | Science › Life sciences | 57.7 | 56.0 | −1.7 |
| LMArena Agent Arena Praise vs Complaint | Tool & computer use › General agents | 29.3 | 32.3 | +1.5 |
| AA-Briefcase v1.1 Elo | Professional › General knowledge work | 1569 | 1544 | −1.4 |
| DAYJOB: Finance (October 2026 rescoring) | Professional › Finance, accounting & tax | 38.5 | 37.3 | −1.2 |
| DAYJOB: Tax | Professional › Finance, accounting & tax | 3.6 | 4.8 | +1.2 |
| GPQA Diamond | Science › Multidisciplinary | 96.8 | 95.8 | −1.0 |
| AA-Omniscience v1.1 Index | Knowledge & reasoning › General knowledge | 46.3 | 44.6 | −0.8 |
| sudo L7 (October 2026 rescoring) | Software › Software engineering | 31.7 | 31.1 | −0.6 |
| GDP.pdf (Artificial Analysis, All-pass) | Professional › General knowledge work | 32.8 | 32.2 | −0.6 |
| LMArena Agent Arena Confirmed Success | Tool & computer use › General agents | 15.7 | 14.9 | −0.4 |
| DrugDiscoveryBench | Science › Life sciences | 69.1 | 68.7 | −0.4 |
| LMArena Agent Arena Tool Hallucination | Tool & computer use › Tool calling | 0.7 | 1.2 | +0.2 |
| ARC-AGI-1 Semi-Private | Knowledge & reasoning › Abstract reasoning | 98.5 | 98.5 | 0.0 |
| OTIS Mock AIME 2024-2025 | Math | 100.0 | 100.0 | 0.0 |
Data: Artificial Analysis, Epoch AI, Surge AI and 19 othersatlas.village.dev