B.O.B. Leaderboard
DeepSeekreleased 24 Apr 2026
DeepSeek V4 Pro
#48 of 9530 of 46 benchmarksBenchmaxxed?
44.6
B.O.B.
Category profile
- Real use (hard to game)2×31
- Reasoning & knowledge56
- Math54
- Coding56
- Agents & real work43
- Independent indexes44
Reality check
Lab tests
52
Real use (8 tests)
31
Scores much higher on lab tests than on real-use tests. It does 21 points better on the usual lab tests than on tests that are hard to train for.
What counts as real usePrice per 1M tokens (OpenRouter)
$1.32 in · $3.96 out
Cost per task on the AA index
$0.67
Output speed (Artificial Analysis)
95 tok/s
§Receipts
Every benchmark result
“Beats” = the share of tracked models this one beat on that test. The raw score is its best reported setting.
| Benchmark | Raw | Beats | Source |
|---|---|---|---|
| Real use (hard to game) · Real use 31 | |||
| LMArena WebDevas deepseek-v4-pro-high-20260813 | 1582 | 65%of 82 | LMArena (blind human votes) |
| LMArena Textas deepseek-v4-pro-high-20260813 | 1464 | 52%of 80 | LMArena (blind human votes) |
| LMArena Agent Arenaas DeepSeek V4 Pro (High) (0813) | 0.004 | 41%of 50 | LMArena (real agent sessions) |
| Kagi LLM Benchmarkas deepseek-v4-pro | 53.5% | 29%of 22 | Kagi |
| Vals Indexas deepseek/deepseek-v4-pro-0813 | 47.6% | 22%of 42 | Vals AI (private test sets) |
| EQ-Bench Creative Writingas deepseek-ai/DeepSeek-V4-Pro | 1553 | 18%of 46 | Sam Paech (open source) |
| SWE-rebenchas DeepSeek-V4 Pro [high] | 41.4% | 15%of 34 | Nebius |
| BridgeBench (overall)as DeepSeek V4 Pro | 460 | 4%of 29 | BridgeMind |
| Reasoning & knowledge · Reasoning 56 | |||
| Mystery Game Puzzlesas deepseek-v4-pro-0813_max | 43.0% | 78%of 57 | Epoch AI Benchmarking Hub |
| Chess Puzzlesas deepseek-v4-pro-0813_max | 46.0% | 70%of 73 | Epoch AI Benchmarking Hub |
| GPQA Diamondas deepseek-v4-pro-0813_max | 91.7% | 70%of 70 | Epoch AI Benchmarking Hub |
| CritPt (physics research)as deepseek-v4-pro-0813_max | 18.0% | 65%of 89 | Epoch AI Benchmarking Hub |
| SimpleQA Verifiedas deepseek-v4-pro-0813_max | 52.9% | 64%of 57 | Epoch AI Benchmarking Hub |
| ARC-AGI-2as deepseek-v4-pro-0813_max | 61.3% | 37%of 44 | Epoch AI Benchmarking Hub |
| ForecastBenchas deepseek-v4-pro_unknown | 55.6% | 6%of 18 | Epoch AI Benchmarking Hub |
| Math · Math 54 | |||
| OTIS Mock AIMEas deepseek-v4-pro-0813_max | 98.6% | 75%of 73 | Epoch AI Benchmarking Hub |
| ProofBenchas deepseek-v4-pro-0813_unknown | 50.0% | 55%of 68 | Epoch AI Benchmarking Hub |
| FrontierMath Tiers 1-3as deepseek-v4-pro-0813_max | 64.6% | 48%of 60 | Epoch AI Benchmarking Hub |
| FrontierMath Tier 4as deepseek-v4-pro-0813_max | 26.8% | 35%of 52 | Epoch AI Benchmarking Hub |
| Coding · Coding 56 | |||
| SWE-bench Verifiedas deepseek-v4-pro_max | 77.6% | 73%of 16 | Epoch AI Benchmarking Hub |
| ALE-Benchas deepseek-v4-pro-0813_max | 1403.2% | 71%of 71 | Epoch AI Benchmarking Hub |
| WeirdMLas deepseek-v4-pro-0813_max | 66.2% | 59%of 52 | Epoch AI Benchmarking Hub |
| SciCodeas deepseek-v4-pro-0813_max | 51.0% | 47%of 86 | Epoch AI Benchmarking Hub |
| FrontierCodeas deepseek-v4-pro-0813_high | 28.5% | 27%of 45 | Epoch AI Benchmarking Hub |
| Agents & real work · Agents 43 | |||
| DTBenchas deepseek-v4-pro-0813_unknown | 93.9% | 64%of 67 | Epoch AI Benchmarking Hub |
| LMCAas deepseek-v4-pro-0813_unknown | 45.5% | 52%of 67 | Epoch AI Benchmarking Hub |
| APEX-Agentsas deepseek-v4-pro-0813_unknown | 47.3% | 38%of 48 | Epoch AI Benchmarking Hub |
| Vending-Bench 2as deepseek-v4-pro_unknown | 3284.5% | 17%of 47 | Epoch AI Benchmarking Hub |
| Independent indexes · Indexes 44 | |||
| Artificial Analysis Intelligence Indexas DeepSeek V4 Pro 0813 (max) | 36 | 51%of 54 | Artificial Analysis |
| Epoch Capabilities Indexas DeepSeek-V4-Pro | 149.2 | 36%of 81 | Epoch AI |