Gemini 4 Hydrogen vs GPT-6 Sol
Gemini 4 Hydrogen leads by 2.1 capability points (≈95% interval −0.3 to +4.6): statistically tied · P(higher) ≈ 95%
Wins 37 of 52 shared benchmarks (14 lost, 1 tied)
Capability vs. relative cost
Gemini 4 HydrogenGPT-6 SolClosed weightsOpen weights
Benchmark wins37 – 14 · 52 shared, 1 tied
Comparing matching setups only (7 benchmarks excluded) · suites counted once
Area← Gemini 4 HydrogenGPT-6 Sol →
Data: Artificial Analysis, Epoch 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 Gemini 4 Hydrogen.
| Humanity's Sixth Sense | Perception & media › Spatial & visual reasoning | 63.4 | 31.2 | −32.2 |
| Terminal-Bench-Science 0.1 | Science › Multidisciplinary | 49.5 | 17.6 | −31.9 |
| Tax Agent Bench | Professional › Finance, accounting & tax | 78.4 | 53.0 | −25.4 |
| Legal Research Bench | Professional › Legal | 53.8 | 28.8 | −25.0 |
| GDP.xlsx | Professional › General knowledge work | 40.9 | 17.1 | −23.8 |
| Terminal-Bench 4.0 (Artificial Analysis) | Software › Software engineering | 53.5 | 30.3 | −23.2 |
| DAYJOB: Finance (October 2026 rescoring) | Professional › Finance, accounting & tax | 38.5 | 19.5 | −19.0 |
| SimpleBench | Knowledge & reasoning › General knowledge | 91.7 | 73.1 | −18.6 |
| AutomationBench 1.0.6 (Zapier) | Tool & computer use › Customer service & operations | 51.1 | 33.2 | −17.9 |
| EBR-bench v4 (card ban, single agent) | Games | 70.5 | 53.3 | −17.1 |
| DAYJOB: Healthcare (October 2026 rescoring) | Health & medicine | 24.0 | 7.2 | −16.8 |
| Professional Reasoning Benchmark - Legal | Professional › Legal | 53.1 | 38.6 | −14.5 |
| TxBench Antibody Discovery V1 (100 evaluations) | Science › Life sciences | 57.7 | 43.3 | −14.3 |
| TaxCalcBench TY25 | Professional › Finance, accounting & tax | 44.0 | 30.0 | −14.0 |
| LMArena Agent Arena Net Improvement (Steering Burden) | Tool & computer use › General agents | 13.5 | 6.5 | −13.8 |
| Humanity's Last Exam Diamond | Knowledge & reasoning › General knowledge | 58.8 | 45.2 | −13.6 |
| Chartography | Perception & media › Documents & charts | 66.6 | 53.6 | −13.0 |
| AutomationBench-AA 1.0.6 | Tool & computer use › Customer service & operations | 74.4 | 61.7 | −12.7 |
| AA-Omniscience v1.1 Index | Knowledge & reasoning › General knowledge | 46.3 | 21.4 | −12.4 |
| AA-Briefcase v1.1 Elo | Professional › General knowledge work | 1569 | 1359 | −11.5 |
| TxBench Oligo Discovery V1 (113 evaluations) | Science › Life sciences | 54.3 | 43.1 | −11.2 |
| SimpleQA Verified (Epoch anti-abstention) | Knowledge & reasoning › General knowledge | 70.9 | 60.7 | −10.2 |
| Gray Swan IPI Attack Success (k=15) | Safety & alignment | 0.0 | 10.1 | −10.1 |
| LMArena Agent Arena Praise vs Complaint | Tool & computer use › General agents | 29.3 | 9.8 | −9.8 |
| Mystery Game Puzzles | Games | 65.0 | 56.0 | −9.0 |
| GDP.pdf (Artificial Analysis, All-pass) | Professional › General knowledge work | 32.8 | 25.2 | −7.6 |
| LMArena Agent Arena Bash Reliability | Tool & computer use › General agents | -4.5 | 10.0 | +7.3 |
| DualEntry Accounting AI Benchmark | Professional › Finance, accounting & tax | 83.2 | 76.2 | −7.0 |
| BuseyBench SVG | Perception & media › Generation | 6.9 | 7.5 | +6.7 |
| Medical Long Context Reasoning (MLCR-AA, October 2026 methodology) | Health & medicine | 27.2 | 33.9 | +6.7 |
| LMArena Agent Arena Confirmed Success | Tool & computer use › General agents | 15.7 | 3.9 | −5.9 |
| LMArena Text Arena Elo | Language & writing › General assistance | 1490 | 1458 | −5.6 |
| Terminal-Bench 4.0 | Software › Software engineering | 54.0 | 49.4 | −4.6 |
| LMArena Agent Arena Steering Burden | Tool & computer use › General agents | 17.4 | 9.3 | −4.1 |
| Chess Puzzles (task version 2.0) | Games | 60.0 | 56.0 | −4.0 |
| FORTRESS Average Risk Score (ARS) | Safety & alignment | 17.5 | 13.5 | +4.0 |
| FrontierCode 2.0 (Artificial Analysis) | Software › Software engineering | 39.0 | 35.6 | −3.4 |
| ARC-AGI-1 Semi-Private | Knowledge & reasoning › Abstract reasoning | 98.5 | 95.5 | −3.0 |
| Artificial Analysis Intelligence Index v5.0.1 | Composite indexes | 41.2 | 44.2 | +2.9 |
| ARC-AGI-2 | Knowledge & reasoning › Abstract reasoning | 92.5 | 89.6 | −2.9 |
| Epoch Capabilities Index (ECI) | Composite indexes | 167 | 164 | −2.4 |
| BigFinanceBench private holdout (final-answer accuracy) | Professional › Finance, accounting & tax | 49.4 | 47.1 | −2.3 |
| FrontierMath Tier 4 (v2) - Private | Math | 87.8 | 90.0 | +2.2 |
| GBENCH One-shot Coding GScore | Games | 66.8 | 68.8 | +2.0 |
| FrontierMath Tiers 1-3 (v2) - Private | Math | 88.1 | 89.8 | +1.8 |
| TxBench Preclinical Pharmacology V1 Clean Reruns | Science › Life sciences | 61.3 | 63.0 | +1.7 |
| SpireClimb | Games | 7.7 | 9.2 | +1.5 |
| Furniture Assembly Benchmark | Perception & media › Spatial & visual reasoning | 57.5 | 58.3 | +0.8 |
| CritPt v2 (agentic, avg@5) | Science › Physical sciences | 80.6 | 80.9 | +0.3 |
| LMArena Agent Arena Tool Hallucination | Tool & computer use › Tool calling | 0.7 | 1.2 | +0.2 |
| Vibe Code Bench v1.1 | Software › Frontend & design | 87.8 | 87.8 | 0.0 |
| OTIS Mock AIME 2024-2025 | Math | 100.0 | 100.0 | 0.0 |
Data: Artificial Analysis, Epoch AI, Arena and 15 othersatlas.village.dev