DeepSeek·released 2025-09-291 source
DeepSeek-V3.2-Exp benchmark scores: 8 benchmarks tracked, leading 1. 0% of its scores are independently reproduced — the rest are self-reported or unverified.
Leads on 1 of 8 benchmarks·best result 83.3 on Fiction.LiveBench (unverified)·0 of 8 independently reproduced
Head-to-headDeepSeek-V3.2-Exp vs Gemini 2.5 Flash (Sep 2025)Leads on: The Agent Company
| Benchmark | Score | Evidence status | Measured |
|---|---|---|---|
| Fiction.LiveBench | 83.3accuracy (%) | unverifiedT1 | 2025-09-29Fiction.live leaderboard |
| DTBench | 79.55accuracy (%) | unverifiedT1 | 2025-09-29https://conceptualreasoning.ai/dtbench |
| Aider polyglot | 74.2accuracy (%) | unverified· optimizedT1 | 2025-09-29Aider LLM Leaderboards |
| The Agent Company | 42.9accuracy (%) | unverifiedT1 | 2025-09-29TheAgentCompany leaderboard |
| WeirdML | 39.48accuracy (%) | unverifiedT1 | 2025-09-29https://htihle.github.io/weirdml.html |
| LMCA | 34.25accuracy (%) | unverifiedT1 | 2025-09-29https://conceptualreasoning.ai/lmca |
| CL-bench | 13.2accuracy (%) | unverifiedT2 | 2025-09-29 |
| CL-bench Life | 9.5accuracy (%) | unverifiedT2 | 2025-09-29 |
Claims drawn from cited facts, not live model generation.
This model was developed by DeepSeek and released in 2025, making it a recent-generation entry from that developer. Its 2025 vintage means it reflects the training methods and tooling of that period rather than older architectures, so its capabilities should be read as contemporary rather than legacy.
3 cited facts
This model is scored on 8 tracked benchmarks, and it currently holds the top spot on 1 of them. Its benchmark record is unverified: no outside evaluation has either confirmed or disputed the reported results, so they should be read as neither independently backed nor contradicted.
3 cited facts
The model trails the state-of-the-art leader by an average of 20.82 points across its benchmarks, a wide gap that places it well behind the front of the pack under the disclosed harness. It holds the top score on only 1 benchmark, so that lone leadership does not offset the overall distance from the leaders.
2 cited facts
5 facts cross-checked across data sources: 5 single-source.
Source facts, citations, and refresh stamp for this record.
Sources: epoch_benchmarks, openrouter_models
DeepSeek-V3.2-Exp is an AI model developed by DeepSeek, released 2025-09-29. Its benchmark record below tags every score as independently reproduced, self-reported, or unverified.
DeepSeek-V3.2-Exp has recorded scores on 8 benchmarks, each shown with its evidence status.
DeepSeek-V3.2-Exp has recorded scores on 8 benchmarks — DTBench, WeirdML, LMCA, Aider polyglot, The Agent Company, Fiction.LiveBench, and 2 more. The full table above shows each score with its evidence status.
0 of 8 recorded scores (0%) are independently reproduced rather than self-reported by the lab.
DeepSeek-V3.2-Exp has 8 tracked claims: 8 unverified.