DeepSeekDeepSeek V3.2
DeepSeek V3.2 is a DeepSeek open-weights reasoning model, released 1 Dec 2025. Its best configuration ranks #197 of 785 on the BenchLeader Index at 56.6 ±6.2, in the upper half. The ± is the point: 311 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 (64) and lowest in coding (51). At $0.315 per million tokens blended it is cheaper than most ranked models. Output speed of 17 tokens per second puts it in the slowest quarter, with a first token in 2.1 s. It has been measured at 4 reasoning-effort settings; this summary describes the best-scoring one, and the tabs above switch between them. Last measured 23 Oct 2026.
- Blended price
- $0.315/M
- $0.280 in · $0.420 out
- Output speed
- 17 tok/s
- OpenRouter traffic, 7-day median; not yet measured by Artificial Analysis
- First answer
- 2.06 s
- Context
- 131k
- Full answer
- –
- Cost per run
- –
- Released
- 1 Dec 2025
- Overall index57
- Reasoning58
- Coding51
- Agents & tools64
- Maths60
- Human preference60
Versions
DeepSeek has shipped 2 models under this name. Each is ranked on its own results; a newer version often has fewer results so far, which holds its index nearer the average until more arrive.
| Model | Released | Index | Rank |
|---|---|---|---|
| DeepSeek V3.2this page | 1 Dec 2025 | 56.6 | #197 |
| DeepSeek V3.1thinking | 21 Aug 2025 | 48.1 | #435 |
Reasoning-effort configurations
The same model behaves differently depending on how much it is allowed to think. Each row is one setting, scored only on the benchmarks that were run at that setting. “Not stated” collects results from publishers that did not say which setting they used; for a reasoning model that is usually its thinking mode, but we do not assume it. Pick a setting here or at the top of the page to see its price, speed and category scores.
| Effort | Index | Rank | Speed | First answer | Chat reply cost | Categories |
|---|---|---|---|---|---|---|
| no reasoning | 47.3 | #460 | 17 tok/s | 2.06 s | $0.0002 | Agents & tools 56 · Composite 45 · Instruction following 45 · Knowledge 40 · Long context 47 · Reasoning 43 |
| high | – | – | 17 tok/s | 2.06 s | $0.0002 | Knowledge 49 · Maths 60 |
| thinking | 55.5 | #228 | 17 tok/s | 2.06 s | $0.0002 | Agents & tools 66 · Coding 52 · Composite 54 · Human preference 61 · Instruction following 61 · Knowledge 52 · Long context 62 · Maths 51 · Reasoning 51 |
| not statedbest | 56.6 | #197 | 17 tok/s | 2.06 s | $0.0002 | Agents & tools 64 · Coding 51 · Human preference 60 · Maths 60 · Reasoning 58 |
How its index has moved
8 Sept 2026 to 21 Oct 2026- not stated
- thinking
- no reasoning
The 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.
Metadata
| Benchmark | thinking | Source | Trend |
|---|---|---|---|
| AA Openness Indexnot in index | 44.4#28 | Artificial Analysis |
Reasoning
| Benchmark | no reasoning | high | thinking | not stated | Source | Trend |
|---|---|---|---|---|---|---|
| GPQA Diamondeffort not stated | – | – | – | 83.4%#102 | Epoch AI Benchmarking Hub | |
| Arena Hard Prompts | – | – | 1446#114 | 1446#114 | Arena (LMArena) | |
| GPQA Diamond (AA)not in index | 73.8%#263 | – | 84.0%#150 | – | Artificial Analysis | |
| Humanity's Last Exam (AA)not in index | 9.0%#338 | – | 24.6%#187 | – | Artificial Analysis | |
| GPQA Diamond (Vals)not in index | – | 80.3%#65 | – | – | Vals AI | |
| Kagi LLM Benchmarkeffort not stated | – | – | – | 52.2%#76 | Kagi LLM Benchmark | |
| CritPt | 1.4%#211 | – | 2.9%#182 | – | Artificial Analysis | |
| CL-benchnot in index | – | – | 13.2%#20 | – | Epoch AI Benchmarking Hub | |
| CL-bench Lifenot in index | – | – | 9.5%#14 | – | Epoch AI Benchmarking Hub | |
| LMCAnot in index | – | – | 29.1%#141 | – | Epoch AI Benchmarking Hub | |
| DTBenchnot in index | – | – | 87.7%#97 | – | Epoch AI Benchmarking Hub |
Coding
| Benchmark | high | thinking | not stated | Source | Trend |
|---|---|---|---|---|---|
| SciCode | – | 38.9%#171 | – | SciCode | |
| WeirdML | – | 39.5%#106 | – | WeirdML | |
| Arena Coding | – | 1475#111 | 1467#120 | Arena (LMArena) | |
| Arena WebDeveffort not stated | – | – | 1272#122 | Arena (LMArena) | |
| LiveCodeBench | – | 80.7%#63 | – | Vals AI | |
| SWE-bench (Vals)not in index | 67.6%#67 | – | – | Vals AI | |
| Aider Polygloteffort not stated | – | – | 74.2%#6 | Aider polyglot leaderboard | |
| Vibe Code Bench v1.1not in index | 5.1%#102 | – | – | Vals AI | |
| Terminal-Bench 2.1 (AA)not in index | – | 46.8%#125 | – | Artificial Analysis |
Agents & tools
| Benchmark | no reasoning | high | thinking | not stated | Source | Trend |
|---|---|---|---|---|---|---|
| Terminal-Bench Hard | 25.0%#138 | – | 35.6%#74 | – | Artificial Analysis | |
| τ²-Bench Telecom (AA)not in index | 33.9%#243 | – | 90.6%#61 | – | Artificial Analysis | |
| BFCL Overall | – | – | 56.7%#12 | 54.1%#17 | Berkeley Function Calling Leaderboard | |
| APEX-Agents (AA)not in index | – | – | 14.5%#24 | – | Artificial Analysis | |
| Poker Agentnot in index | – | – | 1090.3%#4 | – | Vals AI | |
| Terminal-Bench 2.0 (Vals)not in index | – | 36.0%#41 | – | – | Vals AI | |
| TheAgentCompanynot in indexeffort not stated | – | – | – | 42.9%#1 | TheAgentCompany | |
| Vending-Bench 2not in indexeffort not stated | – | – | – | 1034#55 | Andon Labs |
Maths
| Benchmark | high | thinking | not stated | Source | Trend |
|---|---|---|---|---|---|
| OTIS Mock AIMEeffort not stated | – | – | 87.8%#82 | Epoch AI Benchmarking Hub | |
| AIME (Vals) | 84.6%#39 | – | – | Vals AI | |
| MGSM | – | 86.0%#62 | – | Vals AI | |
| MathArena Apexpossible contamination | – | 0.5%#41 | – | MathArena | |
| Arena Maths | – | 1423#118 | 1410#134 | Arena (LMArena) |
Knowledge
| Benchmark | no reasoning | high | thinking | Source | Trend |
|---|---|---|---|---|---|
| AA-Omniscience | -48.2#372 | – | -22.5#253 | Artificial Analysis | |
| MMLU-Pro | – | – | 84.9%#59 | Vals AI | |
| LegalBench | – | – | 76.1%#107 | Vals AI | |
| CorpFin | – | 51.0%#99 | – | Vals AI | |
| TaxEval | – | 68.2%#96 | – | Vals AI | |
| MedQA | – | 93.9%#20 | – | Vals AI | |
| CaseLaw v2not in index | – | 55.4%#34 | – | Vals AI | |
| AA-Omniscience: accuracynot in index | 22.8%#266 | – | 33.0%#167 | Artificial Analysis | |
| AA-Omniscience: non-hallucinationnot in index | 8.0%#471 | – | 17.3%#319 | Artificial Analysis | |
| Global-MMLU-Lite (AA)not in index | – | – | 86.5%#23 | Artificial Analysis |
Instruction following
| Benchmark | no reasoning | thinking | not stated | Source | Trend |
|---|---|---|---|---|---|
| IFBench | 43.1%#228 | 60.7%#125 | – | Artificial Analysis | |
| Arena Instruction Followingnot in index | – | 1416#113 | 1414#120 | Arena (LMArena) |
Human preference
| Benchmark | thinking | not stated | Source | Trend |
|---|---|---|---|---|
| Arena Text | 1425#117 | 1422#122 | Arena (LMArena) | |
| Arena Creative Writingnot in index | 1390#119 | 1409#92 | Arena (LMArena) | |
| Arena Multi-turnnot in index | 1421#126 | 1428#115 | Arena (LMArena) | |
| Arena Longer Queriesnot in index | 1434#114 | 1441#105 | Arena (LMArena) |
Long context
| Benchmark | no reasoning | thinking | Source | Trend |
|---|---|---|---|---|
| AA-LCR | 44.7%#341 | 73.3%#178 | Artificial Analysis |
Composite
| Benchmark | no reasoning | thinking | not stated | Source | Trend |
|---|---|---|---|---|---|
| Epoch Capabilities Indexnot in indexeffort not stated | – | – | 144.9#86 | Epoch AI Benchmarking Hub | |
| AA Intelligence Index v4.3.2 | 13.9#305 | 21.5#212 | – | Artificial Analysis |
Where to run it
Every provider serving this model through OpenRouter, with throughput and first-token latency measured on live traffic over the last 30 minutes and each provider’s own price. Purple marks the best in each column.
| Provider | Speed | First token | Input $/M | Output $/M | Context | Quantisation |
|---|---|---|---|---|---|---|
| SiliconFlow | 18 tok/s | 2.16 s | $0.270 | $0.410 | 164k | fp8 |
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 | $0.0002 | – | 19.7 s |
| Summarize a 30-page report | 12,000 / 600 | $0.0036 | – | 37.4 s |
| Code edit | 6,000 / 1,500 | $0.0023 | – | 1.5 min |
| Agentic coding session | 60,000 / 4,000 | $0.018 | – | 4.0 min |
| Structured extraction | 2,000 / 200 | $0.0006 | – | 13.8 s |
More from DeepSeek
See also
Questions people ask
- What are DeepSeek V3.2's benchmark scores?
- As of 25 Oct 2026, DeepSeek V3.2 scores 56.6 ±6.2 on the BenchLeader Index, strongest in agents & tools (64) and weakest in coding (51). That rests on 10 independent benchmark results, and every score on this page links to its publisher.
- How does DeepSeek V3.2 rank?
- DeepSeek V3.2 is #197 of 785 ranked configurations on 25 Oct 2026. 311 other configurations score within its margin, so the exact position matters less than the band. The leader is Claude Opus 5.5 (max effort) at 72.6.
- How much does DeepSeek V3.2 cost per 1M tokens?
- As of 25 Oct 2026, DeepSeek V3.2 costs $0.280 per 1M input tokens and $0.420 per 1M output tokens, $0.315 blended (three input tokens per output token).
- When was DeepSeek V3.2 released?
- DeepSeek released DeepSeek V3.2 on 1 Dec 2025. BenchLeader has results for it at 4 reasoning-effort settings.
Data as of 25 Oct 2026. Compare these configurations.
Cite as: BenchLeader, “DeepSeek V3.2: benchmarks, pricing, speed and rank”, data as of 25 Oct 2026, https://www.benchleader.com/models/deepseek-v3-2.