Comparing 2 models
DeepSeek V4.1 Flash vs Gemini 3.1 Pro Preview
Benchmarks through
Verdict
DeepSeek V4.1 Flash places higher on 1 of the 1 tasks where both are ranked. DeepSeek V4.1 Flash costs 55.6x less per input token. Only DeepSeek V4.1 Flash has open weights for self-hosting.
- Choose DeepSeek V4.1 Flash for Fast Summaries (#3), you need to self-host.
Built from the published task rankings below; ranks come from each task's own weighted benchmarks, so a win on one task says nothing about another.
Measured head to head
Benchmarks both models were run on
| Evaluation | DeepSeek V4.1 Flash | Gemini 3.1 Pro Preview | Source |
|---|---|---|---|
| FrontierMath Tiers 1-3 (Epoch AI) | 67.4% ± 2.8 | 59.6% ± 2.9 | source · 2026-10-08 |
| GPQA Diamond (Epoch AI) | 89.8% ± 1.7 | 94.1% ± 1.7 | source · 2026-10-09 |
| LiveBench Agentic Coding | 77.3% | 44.1% | source · 2026-10-16 |
| LiveBench Coding | 80.0% | 76.5% | source · 2026-10-16 |
| LiveBench Data Analysis | 79.3% | 78.5% | source · 2026-10-16 |
| LiveBench Instruction Following | 70.0% | 79.1% | source · 2026-10-16 |
| LiveBench Language | 81.2% | 85.4% | source · 2026-10-16 |
| LiveBench Math | 93.3% | 91.0% | source · 2026-10-16 |
| LiveBench Reasoning | 86.7% | 84% | source · 2026-10-16 |
| LMArena Agent | 3 | -3.5 | source · 2026-10-14 |
| LMArena Agent Bash Recovery | 7.9 | -23.3 | source · 2026-10-08 |
| LMArena Agent Steerability | 1.9 | -6.2 | source · 2026-10-14 |
| LMArena Agent Task Outcome | 7.5 | -10.1 | source · 2026-10-14 |
| LMArena Agent Tool Hallucination | 0.5 | -0.2 | source · 2026-10-14 |
| LMArena Creative Writing | 1437 Elo | 1482 Elo | source · 2026-10-15 |
| LMArena Instruction Following | 1473 Elo | 1466 Elo | source · 2026-10-15 |
| LMArena Long Query | 1474 Elo | 1483 Elo | source · 2026-10-15 |
| LMArena Math | 1478 Elo | 1485 Elo | source · 2026-10-15 |
| LMArena Text | 1475 Elo | 1487 Elo | source · 2026-10-15 |
| LMArena Text Factuality | 1468 Elo | 1472 Elo | source · 2026-10-15 |
| LMArena Vision | 1260 Elo | 1280 Elo | source · 2026-10-09 |
| LMArena Vision Diagrams | 1291 Elo | 1309 Elo | source · 2026-10-09 |
| LMArena Vision OCR | 1270 Elo | 1294 Elo | source · 2026-10-09 |
| LMArena WebDev | 1619 Elo | 1447 Elo | source · 2026-10-13 |
| LMArena WebDev Image-to-Web | 1609 Elo | 1481 Elo | source · 2026-10-09 |
| OTIS Mock AIME (Epoch AI) | 98.3% ± 0.6 | 95.6% ± 3.1 | source · 2026-10-08 |
| SimpleQA Verified (Epoch AI) | 42.9% ± 1.6 | 73.5% ± 1.4 | source · 2026-10-13 |
Values are as published by each evaluator, in its own units; rows from different evaluators are not on a common scale.
task evidence
How they place when the question changes
Lower ranks are better. “Best” marks the strongest option in this shortlist for that task.