DeepSeekDeepSeek V3.1 Terminus
DeepSeek V3.1 Terminus is a DeepSeek open-weights reasoning model, released 22 Sept 2025. Its best configuration (thinking reasoning effort) ranks #290 of 777 on the BenchLeader Index at 53.3 ±3.9, in the upper half. The ± is the point: 286 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 long context (60) and lowest in composite (46). At $1.92 per million tokens blended it is pricier than most ranked models. Output speed of 20 tokens per second puts it in the slowest quarter, with a first token in 1.9 s. It has been measured at 3 reasoning-effort settings; this summary describes the best-scoring one, and the tabs above switch between them. Last measured 20 Oct 2026.
- Blended price
- $1.92/M
- $1.64 in · $2.75 out
- Output speed
- 20 tok/s
- OpenRouter traffic, 7-day median; not yet measured by Artificial Analysis
- First answer
- 1.87 s
- Context
- 164k
- Full answer
- –
- Cost per run
- –
- Released
- 22 Sept 2025
- Overall index53
- Reasoning49
- Coding59
- Agents & tools53
- Knowledge50
- Instruction following58
- Human preference59
- Long context60
- Composite46
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 | 47.9 | #438 | 63 tok/s | 1.00 s | $0.0004 | Agents & tools 62 · Composite 45 · Instruction following 44 · Knowledge 42 · Long context 47 · Reasoning 41 |
| thinkingbest | 53.3 | #290 | 20 tok/s | 1.87 s | $0.0015 | Agents & tools 53 · Coding 59 · Composite 46 · Human preference 59 · Instruction following 58 · Knowledge 50 · Long context 60 · Reasoning 49 |
| not stated | 53.2 | #293 | 20 tok/s | 1.87 s | $0.0004 | Coding 47 · Human preference 59 · Maths 57 · Reasoning 56 |
How its index has moved
9 Sept 2026 to 21 Oct 2026- thinking
- not stated
- 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.
Reasoning
| Benchmark | no reasoning | thinking | not stated | Source | Trend |
|---|---|---|---|---|---|
| Arena Hard Prompts | – | 1444#117 | 1424#142 | Arena (LMArena) | |
| GPQA Diamond (AA)not in index | 75.0%#247 | 79.2%#195 | – | Artificial Analysis | |
| Humanity's Last Exam (AA)not in index | 8.7%#341 | 16.4%#237 | – | Artificial Analysis | |
| Kagi LLM Benchmarkeffort not stated | – | – | 57.4%#59 | Kagi LLM Benchmark | |
| CritPt | 0.0%#307 | 1.7%#200 | – | Artificial Analysis | |
| LMCAnot in indexeffort not stated | – | – | 28.6%#143 | Epoch AI Benchmarking Hub | |
| DTBenchnot in indexeffort not stated | – | – | 81.3%#125 | Epoch AI Benchmarking Hub |
Coding
| Benchmark | thinking | not stated | Source | Trend |
|---|---|---|---|---|
| SciCodeeffort not stated | – | 38.0%#179 | SciCode | |
| Arena Coding | 1460#130 | 1440#155 | Arena (LMArena) | |
| SciCode (AA)not in index | 38.0%#185 | – | Artificial Analysis | |
| ALE-Benchnot in indexeffort not stated | – | 745.2#82 | Epoch AI Benchmarking Hub | |
| Terminal-Bench 4.0 (AA)not in index | 0.0%#148 | – | Artificial Analysis | |
| Terminal-Bench 2.1 (AA)not in index | 44.9%#124 | – | Artificial Analysis |
Agents & tools
| Benchmark | no reasoning | thinking | Source | Trend |
|---|---|---|---|---|
| Terminal-Bench Hard | 31.8%#95 | 30.3%#106 | Artificial Analysis | |
| τ²-Bench Telecom (AA)not in index | 37.1%#221 | 37.1%#221 | Artificial Analysis | |
| GDPval-AA v2.1not in index | – | 10.5%#196 | Artificial Analysis | |
| τ³-Banking (AA) | – | 21.0%#89 | Artificial Analysis | |
| AutomationBenchnot in index | – | 2.7%#173 | Artificial Analysis | |
| GDP.pdf (AA)not in index | – | 5.4%#155 | Artificial Analysis |
Maths
| Benchmark | not stated | Source | Trend |
|---|---|---|---|
| Arena Maths | 1397#152 | Arena (LMArena) |
Knowledge
| Benchmark | no reasoning | thinking | Source | Trend |
|---|---|---|---|---|
| AA-Omniscience | -43.5#335 | -26.4#259 | Artificial Analysis | |
| AA-Omniscience: accuracynot in index | 23.5%#250 | 27.7%#197 | Artificial Analysis | |
| AA-Omniscience: non-hallucinationnot in index | 12.4%#372 | 25.2%#241 | Artificial Analysis |
Instruction following
| Benchmark | no reasoning | thinking | not stated | Source | Trend |
|---|---|---|---|---|---|
| IFBench | 41.2%#244 | 57.0%#134 | – | Artificial Analysis | |
| Arena Instruction Followingnot in index | – | 1418#109 | 1394#146 | Arena (LMArena) |
Human preference
| Benchmark | thinking | not stated | Source | Trend |
|---|---|---|---|---|
| Arena Text | 1416#128 | 1415#130 | Arena (LMArena) | |
| Arena Creative Writingnot in index | 1385#129 | 1405#101 | Arena (LMArena) | |
| Arena Multi-turnnot in index | 1415#132 | 1395#159 | Arena (LMArena) | |
| Arena Longer Queriesnot in index | 1445#97 | 1419#132 | Arena (LMArena) |
Long context
| Benchmark | no reasoning | thinking | Source | Trend |
|---|---|---|---|---|
| AA-LCR | 45.3%#330 | 69.3%#221 | Artificial Analysis |
Composite
| Benchmark | no reasoning | thinking | Source | Trend |
|---|---|---|---|---|
| AA Intelligence Index v4.3.2 | 13.9#295 | 14.8#283 | 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 |
|---|---|---|---|---|---|---|
| StreamLake | 45 tok/s | 1.45 s | $0.343 | $1.03 | 128k | – |
| SiliconFlow | 20 tok/s | 1.79 s | $0.270 | $1.00 | 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 | $0.0015 | – | 16.9 s |
| Summarize a 30-page report | 12,000 / 600 | $0.021 | – | 31.9 s |
| Code edit | 6,000 / 1,500 | $0.014 | – | 1.3 min |
| Agentic coding session | 60,000 / 4,000 | $0.109 | – | 3.4 min |
| Structured extraction | 2,000 / 200 | $0.0038 | – | 11.9 s |
More from DeepSeek
See also
Data as of 21 Oct 2026. Compare these configurations.
Cite as: BenchLeader, “DeepSeek V3.1 Terminus: benchmarks, pricing, speed and rank”, https://www.benchleader.com/models/deepseek-v3-1-terminus, data as of 21 Oct 2026.