DeepSeekDeepSeek V3.2 Speciale
DeepSeek V3.2 Speciale is a DeepSeek open-weights reasoning model, released 1 Dec 2025. Its best configuration ranks #297 of 777 on the BenchLeader Index at 53.0 ±4.6, in the upper half. The ± is the point: 294 other configurations score within that range, so they and this one cannot be told apart on quality alone — price and speed are what separate them. It scores highest in agents & tools (65) and lowest in maths (45). At $0.855 per million tokens blended it is mid-priced. Last measured 1 Dec 2025.stale: no new result in six months
- Blended price
- $0.855/M
- $0.580 in · $1.68 out
- Output speed
- –
- First answer
- –
- Context
- 128k
- Full answer
- –
- Cost per run
- –
- Released
- 1 Dec 2025
- Overall index53
- Reasoning53
- Coding48
- Agents & tools65
- Maths45
- Knowledge54
- Instruction following64
- Long context60
- Composite46
How its index has moved
9 Sept 2026 to 18 Oct 2026The index is recomputed from scratch every day, so a line moves when a new benchmark result lands, when a publisher revises a score, or when the models it is normalised against change. Early movement usually means the score is still settling.
Benchmark results
See all LLM benchmarks. One column per reasoning effort. Rank is among every configuration of every model on that benchmark. Hover a score for the run it came from.
Reasoning
| Benchmark | not stated | Source |
|---|---|---|
| SimpleBench | 52.6%#44 | SimpleBench |
| GPQA Diamond (AA)not in index | 87.1%#100 | Artificial Analysis |
| Humanity's Last Exam (AA)not in index | 28.7%#155 | Artificial Analysis |
| CritPt | 7.4%#139 | Artificial Analysis |
Agents & tools
| Benchmark | not stated | Source |
|---|---|---|
| Terminal-Bench Hard | 34.9%#77 | Artificial Analysis |
| τ²-Bench Telecom (AA)not in index | 0.0%#386 | Artificial Analysis |
Maths
| Benchmark | not stated | Source |
|---|---|---|
| MathArena Apexpossible contamination | 9.4%#22 | MathArena |
Knowledge
| Benchmark | not stated | Source |
|---|---|---|
| AA-Omniscience | -17.7#231 | Artificial Analysis |
| AA-Omniscience: accuracynot in index | 37.8%#136 | Artificial Analysis |
| AA-Omniscience: non-hallucinationnot in index | 10.8%#397 | Artificial Analysis |
Instruction following
| Benchmark | not stated | Source |
|---|---|---|
| IFBench | 63.9%#108 | Artificial Analysis |
Long context
| Benchmark | not stated | Source |
|---|---|---|
| AA-LCR | 70.0%#210 | Artificial Analysis |
Composite
| Benchmark | not stated | Source |
|---|---|---|
| AA Intelligence Index v4.3.2 | 14.5#286 | Artificial Analysis |
What a task costs
Estimates from list price, output speed and time to first answer for the best configuration. “With caching” assumes three-quarters of the input is served from the prompt cache. Reasoning tokens are not modelled.
| Workload | Tokens in / out | Cost | With caching | Time |
|---|---|---|---|---|
| Chat reply | 400 / 300 | $0.0007 | – | – |
| Summarize a 30-page report | 12,000 / 600 | $0.0080 | – | – |
| Code edit | 6,000 / 1,500 | $0.0060 | – | – |
| Agentic coding session | 60,000 / 4,000 | $0.042 | – | – |
| Structured extraction | 2,000 / 200 | $0.0015 | – | – |
More from DeepSeek
See also
Data as of 21 Oct 2026. Compare with another model.
Cite as: BenchLeader, “DeepSeek V3.2 Speciale: benchmarks, pricing, speed and rank”, https://www.benchleader.com/models/deepseek-v3-2-speciale, data as of 21 Oct 2026.