DeepSeekDeepSeek V3.2
Best configuration ranks #170 of 610 on the BenchLeader Index at 55.9 ±4.3 (thinking reasoning effort). Last measured 8 Sept 2026. Released 1 Jan 2025.
- Blended price
- $0.315/M
- $0.280 in · $0.420 out
- Output speed
- 30 tok/s
- First answer
- 1.78 s
- Context
- 128k
- Overall index56
- Reasoning62
- Coding49
- Agents & tools66
- Maths41
- Knowledge53
- Instruction following60
- Human preference61
- Long context63
- Composite56
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. “Default” means the publisher did not say which setting was used.
| Effort | Index | Rank | Speed | First answer | Chat reply cost | Categories |
|---|---|---|---|---|---|---|
| high | 50.1 | #294 | 30 tok/s | 1.78 s | $0.0002 | Coding 42 · Knowledge 49 · Maths 60 |
| thinkingbest | 55.9 | #170 | 30 tok/s | 1.78 s | $0.0002 | Agents & tools 66 · Coding 49 · Composite 56 · Human preference 61 · Instruction following 60 · Knowledge 53 · Long context 63 · Maths 41 · Reasoning 62 |
| default | 54.7 | #192 | 30 tok/s | 1.78 s | $0.0002 | Agents & tools 64 · Coding 50 · Composite 47 · Human preference 61 · Instruction following 63 · Knowledge 56 · Long context 61 · Maths 45 · Reasoning 54 |
Benchmark results
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 | high | thinking | default | Source | Trend |
|---|---|---|---|---|---|
| SimpleBench | – | – | 52.6%#44 | SimpleBench | |
| LMArena Hard Prompts | – | 1447#96 | 1446#98 | LMArena | |
| GPQA Diamond (AA)not in index | – | 84.0%#149 | 87.1%#102 | Artificial Analysis | |
| Humanity's Last Exam (AA)not in index | – | 24.6%#136 | 28.7%#113 | Artificial Analysis | |
| GPQA Diamond (Vals)not in index | 80.3%#65 | – | – | Vals AI | |
| Kagi LLM Benchmark | – | – | 52.2%#76 | Kagi LLM Benchmark |
Coding
| Benchmark | high | thinking | default | Source | Trend |
|---|---|---|---|---|---|
| SciCode | – | 38.9%#116 | – | SciCode | |
| WeirdML | – | 39.5%#104 | 46.7%#74 | WeirdML | |
| LMArena Coding | – | 1476#95 | 1465#106 | LMArena | |
| LMArena WebDev | – | 1361#85 | 1272#102 | LMArena | |
| LiveCodeBench | – | 80.7%#63 | – | Vals AI | |
| IOI | 10.7%#33 | – | – | Vals AI | |
| SWE-bench (Vals)not in index | 67.6%#67 | – | – | Vals AI | |
| Aider Polyglot | – | – | 70.2%#9 | Aider polyglot leaderboard |
Agents & tools
| Benchmark | high | thinking | default | Source | Trend |
|---|---|---|---|---|---|
| Terminal-Bench Hard | – | 35.6%#72 | 34.9%#80 | Artificial Analysis | |
| τ²-Bench Telecom (AA)not in index | – | 90.6%#62 | 33.9%#236 | Artificial Analysis | |
| BFCL Overall | – | 56.7%#12 | 54.1%#17 | Berkeley Function Calling Leaderboard |
Maths
| Benchmark | high | thinking | default | Source | Trend |
|---|---|---|---|---|---|
| AIME (Vals) | 84.6%#39 | – | – | Vals AI | |
| MGSM | – | 86.0%#63 | – | Vals AI | |
| MathArena Apex | – | 0.5%#41 | 9.4%#22 | MathArena |
Knowledge
| Benchmark | high | thinking | default | Source | Trend |
|---|---|---|---|---|---|
| AA-Omniscience | – | -22.5#191 | -17.7#174 | Artificial Analysis | |
| MMLU-Pro | – | 84.9%#59 | – | Vals AI | |
| LegalBench | – | 76.1%#102 | – | Vals AI | |
| CorpFin | 51.0%#100 | – | – | Vals AI | |
| TaxEval | 68.2%#97 | – | – | Vals AI | |
| MedQA | 93.9%#20 | – | – | Vals AI |
Instruction following
| Benchmark | high | thinking | default | Source | Trend |
|---|---|---|---|---|---|
| IFBench | – | 60.7%#122 | 63.9%#111 | Artificial Analysis |
Human preference
| Benchmark | high | thinking | default | Source | Trend |
|---|---|---|---|---|---|
| LMArena Text | – | 1425#101 | 1422#105 | LMArena |
Long context
| Benchmark | high | thinking | default | Source | Trend |
|---|---|---|---|---|---|
| AA-LCR | – | 73.3%#129 | 70.0%#161 | Artificial Analysis |
Composite
| Benchmark | high | thinking | default | Source | Trend |
|---|---|---|---|---|---|
| Epoch Capabilities Indexnot in index | – | – | 145.0#72 | Epoch AI Benchmarking Hub | |
| AA Intelligence Index | – | 21.5#159 | 14.5#230 | 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 |
|---|---|---|---|---|---|---|
| AtlasCloud | 31 tok/s | 1.96 s | $0.270 | $0.410 | 164k | fp8 |
| NovitaAI | 30 tok/s | 1.78 s | $0.270 | $0.410 | 164k | fp8 |
| SiliconFlow | 11 tok/s | 1.74 s | $0.270 | $0.410 | 164k | fp8 |
Price history
Listed price per 1M tokens over time, as recorded by OpenRouter for the provider with the longest history.
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 | – | 11.8 s |
| Summarise a 30-page report | 12,000 / 600 | $0.0036 | – | 21.8 s |
| Code edit | 6,000 / 1,500 | $0.0023 | – | 51.8 s |
| Agentic coding session | 60,000 / 4,000 | $0.018 | – | 2.3 min |
| Structured extraction | 2,000 / 200 | $0.0006 | – | 8.4 s |
See also
Data as of 9 Sept 2026. Compare these configurations.