DeepSeek V4.1 Flash 1017 vs Gemini 4 Hydrogen
TryTop 2 now
Gemini 4 Hydrogen leads by 4.9 capability points (≈95% interval +2.3 to +7.6): a clear lead over DeepSeek V4.1 Flash 1017 · P(higher) ≈ 100%
Wins 60 of 63 shared benchmarks (3 lost)
Capability vs. relative cost
DeepSeek V4.1 Flash 1017Gemini 4 HydrogenClosed weightsOpen weights
Benchmark wins3 – 60 · 63 shared
Comparing matching setups only (5 benchmarks excluded) · suites counted once
Area← DeepSeek V4.1 Flash 1017Gemini 4 Hydrogen →
Data: Artificial Analysis, Vals AI, Epoch AI and 14 othersatlas.village.dev
Margins
Each benchmark scaled 0–100% (its declared range, else the field’s range); Δ is Gemini 4 Hydrogen minus DeepSeek V4.1 Flash 1017.
| FrontierMath Tier 4 (v2) - Private | Math | 46.3 | 87.8 | +41.5 |
| Humanity's Last Exam Diamond | Knowledge & reasoning › General knowledge | 20.8 | 58.8 | +38.0 |
| Humanity's Sixth Sense | Perception & media › Spatial & visual reasoning | 30.1 | 63.4 | +33.3 |
| Terminal-Bench-Science 0.1 | Science › Multidisciplinary | 10.5 | 41.4 | +31.0 |
| SimpleQA Verified (Epoch anti-abstention) | Knowledge & reasoning › General knowledge | 42.5 | 70.9 | +28.4 |
| DAYJOB: Accounting | Professional › Finance, accounting & tax | 7.6 | 33.6 | +26.0 |
| DAYJOB: Finance (October 2026 rescoring) | Professional › Finance, accounting & tax | 12.5 | 38.5 | +26.0 |
| CritPt v2 (agentic, avg@5) | Science › Physical sciences | 52.3 | 78.0 | +25.7 |
| Terminal-Bench 4.0 (Artificial Analysis) | Software › Software engineering | 28.8 | 54.0 | +25.3 |
| BuseyBench SVG | Perception & media › Generation | 4.3 | 6.9 | +25.1 |
| LMArena Agent Arena Net Improvement (Steering Burden) | Tool & computer use › General agents | 0.8 | 13.5 | +24.8 |
| Artificial Analysis Engineering Index | Science › Engineering | 37.6 | 60.1 | +22.5 |
| GDP.xlsx | Professional › General knowledge work | 18.4 | 40.9 | +22.5 |
| AA-Omniscience v1.1 Index | Knowledge & reasoning › General knowledge | 1.5 | 46.3 | +22.4 |
| Terminal-Bench 4.0 | Software › Software engineering | 31.8 | 54.0 | +22.2 |
| HANDBOOK.md Agents v1.1 | Tool & computer use › Customer service & operations | 24.0 | 44.6 | +20.6 |
| FrontierCode 2.0 - Pass Rate | Software › Software engineering | 23.0 | 43.2 | +20.2 |
| Artificial Analysis Legal Index | Professional › Legal | 37.7 | 56.7 | +19.0 |
| FrontierCode 2.0 - Score | Software › Software engineering | 20.2 | 38.9 | +18.7 |
| DAYJOB: Healthcare (October 2026 rescoring) | Health & medicine | 6.0 | 24.0 | +18.0 |
| FrontierMath Tiers 1-3 (v2) - Private | Math | 71.2 | 88.1 | +16.8 |
| Artificial Analysis Economics Index | Composite indexes | 38.5 | 55.0 | +16.6 |
| Mystery Game Puzzles | Games | 49.0 | 65.0 | +16.0 |
| Artificial Analysis Finance & Accounting Index | Professional › Finance, accounting & tax | 42.2 | 58.1 | +15.9 |
| GDP.pdf (Artificial Analysis, All-pass) | Professional › General knowledge work | 17.2 | 32.8 | +15.6 |
| Artificial Analysis Coding Agent Index v2.1 | Software › Software engineering | 36.8 | 52.3 | +15.5 |
| LMArena Agent Arena Praise vs Complaint | Tool & computer use › General agents | -1.5 | 29.3 | +15.4 |
| ARC-AGI-2 | Knowledge & reasoning › Abstract reasoning | 78.1 | 92.5 | +14.4 |
| Code Migration v1.1 (Vals Index v2.2 subset) | Software › Software engineering | 68.5 | 82.7 | +14.2 |
| Chess Puzzles (task version 2.0) | Games | 46.0 | 60.0 | +14.0 |
| Artificial Analysis Healthcare & Medical Index | Health & medicine | 41.3 | 53.3 | +12.0 |
| GBENCH One-shot Coding GScore | Games | 55.4 | 66.8 | +11.4 |
| Ramp SWE-Bench v2 | Software › Software engineering | 60.8 | 71.6 | +10.8 |
| CWE-bench v2 Pass@1 | Software › Cybersecurity | 57.6 | 68.1 | +10.5 |
| Bug Hunt Bench | Software › Software engineering | 40.7 | 31.1 | −9.6 |
| Legal Research Bench | Professional › Legal | 45.2 | 53.8 | +8.7 |
| Artificial Analysis Intelligence Index v5.0.1 | Composite indexes | 32.6 | 41.2 | +8.6 |
| Artificial Analysis Strategy & Ops Index | Professional › General knowledge work | 49.2 | 57.7 | +8.5 |
| LMArena Agent Arena Steering Burden | Tool & computer use › General agents | 2.3 | 17.4 | +7.6 |
| Agents' Last Exam v1.1 (ALE v1.1) | Professional › General knowledge work | 36.1 | 43.4 | +7.4 |
| Vals Index v2.2 | Composite indexes | 61.3 | 68.5 | +7.2 |
| DAYJOB: Law (November 2026 release) | Professional › Legal | 2.8 | 9.7 | +6.9 |
| LiveBench 2.0 (2026-11-06 release) | Composite indexes | 70.3 | 77.0 | +6.7 |
| LMArena Agent Arena Confirmed Success | Tool & computer use › General agents | 2.2 | 15.7 | +6.7 |
| Excel Modeling Benchmark v1.1 | Professional › Finance, accounting & tax | 69.2 | 75.8 | +6.6 |
| CursorBench 4.0 | Software › Software engineering | 47.7 | 54.3 | +6.6 |
| AutomationBench-AA 1.0.6 | Tool & computer use › Customer service & operations | 68.5 | 74.4 | +5.9 |
| Epoch Capabilities Index (ECI) | Composite indexes | 159 | 167 | +5.9 |
| LMArena Code Arena WebDev | Software › Frontend & design | 1659 | 1703 | +5.9 |
| Vals Finance Index | Professional › Finance, accounting & tax | 60.9 | 66.1 | +5.2 |
| AA-Briefcase v1.1 Elo | Professional › General knowledge work | 1416 | 1497 | +4.4 |
| ARC-AGI-1 Semi-Private | Knowledge & reasoning › Abstract reasoning | 94.5 | 98.5 | +4.0 |
| Tax Agent Bench | Professional › Finance, accounting & tax | 74.9 | 78.4 | +3.5 |
| HLAB-Vals-1.1 | Professional › Legal | 14.2 | 11.3 | −2.9 |
| LMArena Agent Arena Bash Reliability | Tool & computer use › General agents | 1.1 | -4.5 | −2.8 |
| DAYJOB: Tax | Professional › Finance, accounting & tax | 1.2 | 3.6 | +2.4 |
| Finance Agent v2.1 | Professional › Finance, accounting & tax | 62.9 | 65.1 | +2.2 |
| Vals Legal Index | Professional › Legal | 71.9 | 73.6 | +1.8 |
| DualEntry Accounting AI Benchmark | Professional › Finance, accounting & tax | 82.2 | 83.2 | +1.0 |
| OTIS Mock AIME 2024-2025 | Math | 99.4 | 100.0 | +0.6 |
| Medical Long Context Reasoning (MLCR-AA, October 2026 methodology) | Health & medicine | 26.7 | 27.2 | +0.6 |
| LMArena Agent Arena Tool Hallucination | Tool & computer use › Tool calling | -0.1 | 0.7 | +0.4 |
| Vibe Code Bench v1.1 | Software › Frontend & design | 87.7 | 87.8 | 0.0 |
Data: Artificial Analysis, Vals AI, Epoch AI and 14 othersatlas.village.dev