Capabilities
- AIME 2025 96
- HMMT 2025 99.2
- HMMT Nov 2025 94.4
- IMO-AnswerBench 84.5
- LiveCodeBench 88.7
- Codeforces rating 2701
- GPQA Diamond 85.7
- Humanity's Last Exam 30.6
- IMO 2025 35
- CMO 2025 102
- IOI 2025 492
- ICPC World Finals 2025 10
DeepSeek-V3.2-Speciale
DeepSeek-V3.2-Speciale ranks first on CMIMC 2025.
25
Benchmarks with a score
1
Ranked first, out of five or more models
2
In the top three
DeepSeek Dec 2025 DeepSeek-V3.2-Speciale
Open weights
148.7
#105 of 404
Self-Reported Index, from the lab's own numbers. No Epoch Capabilities Index score yet.
- MathArena (Apex, IMO, USAMO) 9.38% #22 /47
- General365 57.5% #7 /23
- AIME 2025 96% #21 /143
Where it ranks
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.
Knowledge and language
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| Humanity's Last Exam | 30.6% | #79 of 154 | Lab model card | |
| Creative Writing v3 | 60.0 | #83 of 120 | Official leaderboard | |
| Longform Creative Writing (EQ-Bench) | 48.8 | #84 of 119 | Official leaderboard |
Math and science
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| CMIMC 2025 | 95.6% | #1 of 33 | Official leaderboard | |
| BRUMO 2025 | 99.2% | #3 of 42 | Official leaderboard | |
| HMMT 2025 | 99.2% | #4 of 54 | Lab model card | |
| AIME 2025 | 96% | #21 of 143 | Lab model card | |
| GPQA Diamond | 85.7% | #72 of 302 | Lab model card | |
| Putnam 2025 | 103.0 | #1 of 4 | Official leaderboard | |
| HMMT Nov 2025 | 93.3% | #6 of 22 | Official leaderboard | |
| SMT 2025 | 89.2% | #12 of 40 | Official leaderboard | |
| General365 | 57.5% | #7 of 23 | Official leaderboard | |
| IMO-AnswerBench | 84.5% | #14 of 37 | Lab model card | |
| MathArena Apex Shortlist | 69.7% | #16 of 38 | Official leaderboard | |
| MathArena (Apex, IMO, USAMO) | 9.38% | #22 of 47 | Official leaderboard | |
| SimpleBench | 52.6% | #57 of 106 | Official leaderboard | |
| CMO 2025 | 102.0 | #3 of 3 | Lab model card |
Coding
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| LiveCodeBench | 88.7% | #4 of 54 | Lab model card | |
| Codeforces rating | 2,701 | #9 of 47 | Lab model card | |
| MLE-bench | 56.4% | #4 of 12 | Official leaderboard | |
| ICPC World Finals 2025 | 10.0 | #1 of 3 | Lab model card | |
| WeirdML | 46.7% | #54 of 112 | Official leaderboard | |
| IOI 2025 | 492.0 | #2 of 3 | Lab model card |
Agents and tools
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| LisanBench | 5072.7 | #24 of 133 | Official leaderboard | |
| SnakeBench | 9 | #183 of 188 | Official leaderboard |
Its lab's numbers put it at 148.7, number 105 of 404
The Self-Reported Index fits one score to the 7 capability results DeepSeek reported for this model, on the scale of the Epoch Capabilities Index. Labs choose which benchmarks to show and how to run them, so read it as the lab's claim, not an independent measurement. The 90% interval runs from 145.0 to 152.4. How we compute it.
- HMMT 2025 99.2
- AIME 2025 96
- HMMT Nov 2025 94.4
- LiveCodeBench 88.7
- GPQA Diamond 85.7
- IMO-AnswerBench 84.5
- Humanity's Last Exam 30.6
Its lab's numbers next to 3 leaderboards
Where the benchmark's own leaderboard, or Epoch AI's run, also has this model. Gaps mostly come from different harnesses, effort levels and task subsets. Every pair.
| Benchmark | Lab says | Board says | Gap |
|---|---|---|---|
| HMMT 2025 | 99.2% HMMT Feb 2025 (Pass@1), DeepSeek-V3.2-Speciale (Table 3) | 97.5% Leaderboard | +1.7 pts |
| HMMT Nov 2025 | 94.4% HMMT Nov 2025 (Pass@1), DeepSeek-V3.2-Speciale (Table 3) | 93.3% Leaderboard | +1.1 pts |
| AIME 2025 | 96% AIME 2025 (Pass@1), 23k output tokens, DeepSeek-V3.2-Speciale (Table 3) | 95.8% Leaderboard | +0.2 pts |
Its system card
Every result in the longer documents the lab published with it. 12 of them aren't in the launch post.