DeepSeek V4.1 Flash 1017 ranks first on 2 of the 54 benchmarks it has scores on.
DeepSeek Oct 2026 Open weights Mid-size
Its best score on each benchmark, and where that ranks among every model with a score there. We rank each model once, at its best setting.
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| MultiChallenge | 74.9% | #5 of 35 | Official leaderboard | |
| Extended NYT Connections | 92.8% | #14 of 97 | Official leaderboard | |
| Humanity's Last Exam | 42.6% | #55 of 147 | Model card | |
| SimpleQA Verified | 42.5% | #45 of 84 | Epoch AI |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| OTIS Mock AIME 2024-2025 | 99.4% | #17 of 202 | Epoch AI | |
| GPQA Diamond | 93.2% | #20 of 219 | Epoch AI | |
| CritPt | 24% | #22 of 160 | Epoch AI | |
| ARC-AGI-2 | 78.1% | #18 of 95 | Official leaderboard | |
| ARC-AGI | 94.5% | #22 of 92 | Official leaderboard | |
| FrontierMath Tiers 1-3 v2 | 71.2% | #27 of 89 | Epoch AI | |
| FrontierMath Tier 4 v2 | 46.3% | #25 of 71 | Epoch AI | |
| MathArena (Apex, IMO, USAMO) | 72.1% | #5 of 13 | Model card | |
| CritPt v2 | 50% | #10 of 11 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| NL2Repo-Bench | 68.2% | #1 of 28 | Model card | |
| DeepSWE | 77% | #4 of 51 | Model card | |
| Vibe Code Bench | 87.7% | #12 of 115 | Official leaderboard | |
| ALE-Bench | 1,622 | #18 of 121 | Epoch AI | |
| Tencent WorkBuddy Bench | 78.3% | #2 of 10 | Official leaderboard | |
| Code Migration v1.1 | 67.5% | #7 of 29 | Official leaderboard | |
| Code Migration | 44.2% | #25 of 78 | Official leaderboard | |
| CursorBench | 47.7% | #7 of 19 | Epoch AI | |
| WeirdML v3 | 9.19% | #14 of 25 | Official leaderboard | |
| FrontierCS 2 | 16.7% | #11 of 16 | Official leaderboard | |
| FrontierCode 2.0 | 19.3% | #18 of 26 | Official leaderboard | |
| ProgramBench | 29.1% | #16 of 20 | Model card |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| LisanBench | 21107.3 | #4 of 133 | Official leaderboard | |
| AutomationBench | 60.5% | #3 of 41 | Model card | |
| Mystery Game Puzzles | 49% | #14 of 80 | Epoch AI | |
| Chess Puzzles | 46% | #26 of 83 | Epoch AI | |
| Terminal-Bench 4.0 | 42.3% | #12 of 23 | Model card |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| APEX-Agents | 69.8% | #4 of 56 | Epoch AI | |
| Tax Agent Bench | 44.4% | #7 of 79 | Official leaderboard | |
| Vals Finance Agent v2 | 57% | #17 of 80 | Official leaderboard | |
| Vals Finance Index | 60.9% | #8 of 35 | Official leaderboard | |
| Legal Research Bench | 45.2% | #21 of 88 | Official leaderboard | |
| Harvey LAB (Legal Agent Benchmark) | 9.17% | #24 of 80 | Official leaderboard | |
| Vals Legal Index | 71.9% | #10 of 33 | Official leaderboard | |
| Vals Index v2.2 | 61.3% | #10 of 32 | Official leaderboard | |
| Harvey LAB v1.1 (HLAB-Vals-1.1) | 14.2% | #11 of 33 | Official leaderboard | |
| Agents' Last Exam | 35.4% | #9 of 25 | Model card | |
| Vals Finance Agent v2.1 | 62.9% | #13 of 35 | Official leaderboard | |
| Excel Modeling Benchmark | 61.8% | #32 of 76 | Official leaderboard | |
| PRBench | 47.3% | #21 of 44 | Official leaderboard | |
| Excel Modeling Benchmark v1.1 | 69.2% | #18 of 37 | Official leaderboard | |
| FoodTruck Bench | $53,289 | #7 of 10 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| ZeroBench | 51% | #1 of 25 | Model card | |
| BabyVision | 92.5% | #2 of 20 | Model card | |
| VoxelBench | 2,047 | #11 of 56 | Official leaderboard | |
| Chartography | 83% | #3 of 12 | Model card | |
| Humanity's Sixth Sense | 30.1% | #17 of 29 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| DrugDiscoveryBench | 60.6% | #4 of 19 | Official leaderboard | |
| DeepsecBench | 30.0 | #8 of 28 | Official leaderboard | |
| FORTRESS | 27.2% | #37 of 66 | Official leaderboard |
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| LiveBench | 72.9% | #9 of 28 | Official leaderboard |
DeepSeek's models in color, against every other model on the Epoch Capabilities Index, by release date.
The benchmarks the lab chose to headline, with the scores it gave.