Gemini 4 Argon vs Gemini 4 Hydrogen
TryTop 2 now
Gemini 4 Hydrogen leads by 0.1 capability points (≈95% interval −2.5 to +2.6): statistically tied · P(higher) ≈ 52%
Wins 55 of 85 shared benchmarks (27 lost, 3 tied)
Capability vs. relative cost
Gemini 4 ArgonGemini 4 HydrogenClosed weightsOpen weights
Benchmark wins27 – 55 · 85 shared, 3 tied
Comparing matching setups only (1 benchmark excluded) · suites counted once
Area← Gemini 4 ArgonGemini 4 Hydrogen →
Data: Artificial Analysis, Surge AI, Epoch AI and 24 othersatlas.village.dev
Margins
Each benchmark scaled 0–100% (its declared range, else the field’s range); Δ is Gemini 4 Hydrogen minus Gemini 4 Argon.
| Terminal-Bench-Science 0.1 | Science › Multidisciplinary | 48.6 | 15.2 | −33.3 |
| TxBench-Antibody-Development V1 (60 evaluations) | Science › Life sciences | 31.1 | 60.6 | +29.5 |
| Terminal-Bench 4.0 (Artificial Analysis) | Software › Software engineering | 55.1 | 29.3 | −25.8 |
| Humanity's Last Exam Diamond | Knowledge & reasoning › General knowledge | 42.5 | 58.8 | +16.3 |
| Ramp SWE-Bench v2 (Bash harness) | Software › Software engineering | 54.8 | 70.8 | +16.0 |
| HLAB-Vals-1.1 | Professional › Legal | 27.1 | 11.3 | −15.8 |
| DAYJOB: Accounting | Professional › Finance, accounting & tax | 18.0 | 33.6 | +15.6 |
| sudo L7 (October 2026 rescoring) | Software › Software engineering | 16.1 | 31.7 | +15.6 |
| KradleBench | Games | 28.6 | 43.9 | +15.3 |
| Furniture Assembly Benchmark | Perception & media › Spatial & visual reasoning | 72.5 | 57.5 | −15.0 |
| Medical Long Context Reasoning (MLCR-AA, October 2026 methodology) | Health & medicine | 13.9 | 27.2 | +13.3 |
| FrontierCode 2.0 - Pass Rate | Software › Software engineering | 30.6 | 43.2 | +12.6 |
| Bug Hunt Bench | Software › Software engineering | 38.7 | 27.0 | −11.7 |
| Dual-Arm Sim (task completion) | Tool & computer use › Robotics | 11.3 | 0.0 | −11.2 |
| FrontierCode 2.0 - Score | Software › Software engineering | 27.6 | 38.9 | +11.2 |
| FrontierMath Tier 4 (v2) - Private | Math | 78.0 | 87.8 | +9.8 |
| Artificial Analysis Intelligence Index v5.0.1 | Composite indexes | 50.4 | 41.2 | −9.2 |
| Code Migration v1.1 (Vals Index v2.2 subset) | Software › Software engineering | 74.4 | 82.7 | +8.2 |
| DAYJOB: Finance (October 2026 rescoring) | Professional › Finance, accounting & tax | 30.5 | 38.5 | +8.0 |
| HANDBOOK.md Agents v1.1 | Tool & computer use › Customer service & operations | 36.6 | 44.6 | +8.0 |
| Humanity's Sixth Sense | Perception & media › Spatial & visual reasoning | 55.4 | 63.4 | +8.0 |
| DualEntry Accounting AI Benchmark | Professional › Finance, accounting & tax | 91.1 | 83.2 | −7.9 |
| GDP.pdf (Artificial Analysis, All-pass) | Professional › General knowledge work | 25.6 | 32.8 | +7.2 |
| BuseyBench SVG | Perception & media › Generation | 7.6 | 6.9 | −7.0 |
| Artificial Analysis Coding Agent Index v2.1 | Software › Software engineering | 45.5 | 52.3 | +6.8 |
| HiL-Bench (Human-in-Loop Benchmark) | Tool & computer use › General agents | 62.0 | 55.3 | −6.7 |
| LMArena Agent Arena Net Improvement (Steering Burden) | Tool & computer use › General agents | 10.3 | 13.5 | +6.3 |
| TxBench Antibody Discovery V1 (100 evaluations) | Science › Life sciences | 63.7 | 57.7 | −6.0 |
| EBR-bench v4 (card ban, single agent) | Games | 64.8 | 70.5 | +5.7 |
| SWE-Marathon v1.1 | Software › Software engineering | 43.8 | 49.4 | +5.6 |
| LMArena Text Arena Elo | Language & writing › General assistance | 1521 | 1490 | −5.4 |
| BigFinanceBench private holdout (final-answer accuracy) | Professional › Finance, accounting & tax | 54.7 | 49.4 | −5.3 |
| CritPt v2 (agentic, avg@5) | Science › Physical sciences | 72.9 | 78.0 | +5.1 |
| SimpleQA Verified (Epoch anti-abstention) | Knowledge & reasoning › General knowledge | 76.0 | 70.9 | −5.1 |
| Chartography | Perception & media › Documents & charts | 71.6 | 66.6 | −5.0 |
| CursorBench 4.1 | Software › Software engineering | 56.5 | 61.4 | +4.9 |
| GradeBench | Professional › Education | 56.5 | 51.9 | −4.6 |
| DAYJOB: Law (November 2026 release) | Professional › Legal | 5.9 | 9.7 | +3.8 |
| DrugDiscoveryBench | Science › Life sciences | 65.4 | 69.1 | +3.7 |
| Professional Reasoning Benchmark - Legal | Professional › Legal | 49.4 | 53.1 | +3.7 |
| SWE-Marathon v1.2 | Software › Software engineering | 54.5 | 58.1 | +3.7 |
| Terminal-Bench 4.0 | Software › Software engineering | 57.6 | 54.0 | −3.5 |
| ComplexConstraints (November 2026 rescoring) | Language & writing › Instruction following | 70.3 | 73.7 | +3.4 |
| TxBench Preclinical Pharmacology V1 Clean Reruns | Science › Life sciences | 64.7 | 61.3 | −3.3 |
| Agents' Last Exam v1.1 (ALE v1.1) | Professional › General knowledge work | 40.2 | 43.4 | +3.3 |
| TxBench Oligo Discovery V1 (113 evaluations) | Science › Life sciences | 51.0 | 54.3 | +3.3 |
| Vals Legal Index | Professional › Legal | 76.8 | 73.6 | −3.1 |
| FrontierSWE V2 Best@5 | Software › Software engineering | 64.3 | 61.2 | −3.1 |
| CWE-bench v2 Pass@1 | Software › Cybersecurity | 65.0 | 68.1 | +3.1 |
| AutomationBench-AA 1.0.6 | Tool & computer use › Customer service & operations | 77.5 | 74.4 | −3.1 |
| LMArena Code Arena WebDev | Software › Frontend & design | 1681 | 1703 | +3.0 |
| OSWorld 2.1 Offline, 500-step Partial Score | Tool & computer use › Computer & browser use | 74.7 | 77.6 | +2.9 |
| GBENCH One-shot Coding GScore | Games | 69.7 | 66.8 | −2.8 |
| micro1 PharmaBench | Professional › General knowledge work | 8.6 | 11.4 | +2.8 |
| LMArena Agent Arena Praise vs Complaint | Tool & computer use › General agents | 23.8 | 29.3 | +2.8 |
| LMArena Agent Arena Steering Burden | Tool & computer use › General agents | 12.2 | 17.4 | +2.6 |
| GDP.xlsx | Professional › General knowledge work | 38.3 | 40.9 | +2.6 |
| Finance Agent v2.1 | Professional › Finance, accounting & tax | 67.7 | 65.1 | −2.5 |
| Professional Reasoning Benchmark - Finance | Professional › Finance, accounting & tax | 49.9 | 52.3 | +2.4 |
| FrontierMath Tiers 1-3 (v2) - Private | Math | 86.0 | 88.1 | +2.1 |
| MultiChallenge | Language & writing › Conversation | 78.7 | 80.6 | +1.9 |
| LMArena Agent Arena Confirmed Success | Tool & computer use › General agents | 12.6 | 15.7 | +1.5 |
| CursorBench 4.0 | Software › Software engineering | 52.9 | 54.3 | +1.4 |
| ARC-AGI-2 | Knowledge & reasoning › Abstract reasoning | 91.3 | 92.5 | +1.3 |
| AutomationBench 1.0.6 (Zapier) | Tool & computer use › Customer service & operations | 50.1 | 51.1 | +1.1 |
| Chess Puzzles (task version 2.0) | Games | 59.0 | 60.0 | +1.0 |
| Mystery Game Puzzles | Games | 64.0 | 65.0 | +1.0 |
| Excel Modeling Benchmark v1.1 | Professional › Finance, accounting & tax | 74.8 | 75.8 | +1.0 |
| Legal Research Bench | Professional › Legal | 54.8 | 53.8 | −1.0 |
| FrontierSWE V2 Worst@5 | Software › Software engineering | 44.5 | 45.3 | +0.9 |
| LMArena Agent Arena Bash Reliability | Tool & computer use › General agents | -6.0 | -4.5 | +0.7 |
| LiveBench 2.0 (2026-11-06 release) | Composite indexes | 77.7 | 77.0 | −0.7 |
| AA-Briefcase v1.1 Elo | Professional › General knowledge work | 1485 | 1497 | +0.6 |
| GDP.pdf (November 2026 rescoring) | Professional › General knowledge work | 53.6 | 54.2 | +0.6 |
| Tax Agent Bench | Professional › Finance, accounting & tax | 79.0 | 78.4 | −0.6 |
| DAYJOB: Tax | Professional › Finance, accounting & tax | 3.2 | 3.6 | +0.4 |
| DAYJOB: Healthcare (October 2026 rescoring) | Health & medicine | 23.6 | 24.0 | +0.4 |
| GPQA Diamond | Science › Multidisciplinary | 96.4 | 96.8 | +0.4 |
| Epoch Capabilities Index (ECI) | Composite indexes | 167 | 167 | +0.2 |
| Vals Finance Index | Professional › Finance, accounting & tax | 66.0 | 66.1 | +0.1 |
| AA-Omniscience v1.1 Index | Knowledge & reasoning › General knowledge | 46.1 | 46.3 | +0.1 |
| LMArena Agent Arena Tool Hallucination | Tool & computer use › Tool calling | 0.8 | 0.7 | 0.0 |
| 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 |
| SimpleBench | Knowledge & reasoning › General knowledge | 91.7 | 91.7 | 0.0 |
Data: Artificial Analysis, Surge AI, Epoch AI and 24 othersatlas.village.dev