Inference
EN
DeepSeek

DeepSeek V3.1

DeepSeek V3.1 is a standard model from DeepSeek. 14 benchmarks count toward its index, in 6 categories.

  • Open weights
  • Standard
  • Input: text
  • Released 2025-08
RankedEnough results for an index.
IndexAll categories, out of 100.37.9±5.8Median 51.0
CoverageIndex weight that has results.85%
SpeedOutput tokens per second.98/sMedian 67/s
First tokenSeconds to the first output token.1.4 sMedian 1.3 s
Input / 1MUS dollars per 1M input tokens.—Median $0.60
Output / 1MUS dollars per 1M output tokens.—Median $2.55
ContextMaximum tokens in one request.128KMedian 262K
EloLMArena rating and rank.1420±6 #118Median 1422

Run locally

684.53B parameters
16-bit
1438.5 GB
8-bit
764.7 GB
4-bit
436.7 GB
Weights
Hugging Face (opens in a new tab) ↗

Memory need with a short context. The 4-bit copy needs more than 128 GB. Compare local models

Capabilities

Index per category, out of 100.
Out of 100
AgenticMulti-step tasks with tools.3 counted38.0182 / 218
CodingCode writing and repair.2 counted43.0172 / 209
ReasoningLogic problems and puzzles.2 counted38.2173 / 227
MultimodalTasks with images and text.None counted——
KnowledgeFacts and expert knowledge.4 counted41.5183 / 268
MultilingualTasks in many languages.None counted——
InstructionTasks with strict prompt rules.1 counted24.3—
MathMath problems.2 counted29.9144 / 168

Results

14 counted
BenchmarkThe benchmark name.ResultThe score from the source.PlacePlace among results on this benchmark.
Massive Multitask Language Understanding ProfessionalKnowledge83.7%28 / 75
Terminal-Bench 1.0Agentic41.3% ±5.518 / 47
Graduate-Level Google-Proof Q&AKnowledge74.9%82 / 141
Artificial Analysis GPQA DiamondKnowledge73.5%124 / 199
LiveCodeBench: Holistic and Contamination Free Evaluation of Large Language Models for CodeCoding56.4%8 / 12
Software Engineering Benchmark VerifiedCoding66.0%79 / 114
Critical Physics TasksReasoning0.0%150 / 214
Artificial Analysis Omniscience AccuracyKnowledge23.1%115 / 164
τ²-Bench Tool-Agent-User EvaluationAgentic34.8%109 / 149
American Invitational Mathematics Examination 2024Math66.3%20 / 27
Artificial Analysis Long Context ReasoningReasoning47.0%161 / 217
Artificial Analysis Humanity's Last ExamKnowledge6.7%165 / 222
Artificial Analysis IFBenchInstruction37.8%115 / 146
American Invitational Mathematics Examination 2025Math49.8%42 / 53
Terminal-Bench 2.0 · Vals AIAgentic22.5% ±4.456 / 66
CaseLaw v2Legal53.9% ±0.638 / 53
CorpFin v2Finance51.5% ±1.0103 / 131

Sources

BenchLMCC BY-NC 4.0Vals AINo licence stated

History

5 releases
Index38.8 → 37.9
23 Oct 202612 Nov 2026

Questions

How good is DeepSeek V3.1?
Its index is 37.9 out of 100. That is place 222 of 283 ranked models.
What is DeepSeek V3.1 best at?
Knowledge: 41.5 out of 100, place 183 of 268.
How fast is DeepSeek V3.1?
98 output tokens per second, 1.4 s to the first token.
What is the context window of DeepSeek V3.1?
128K tokens.
Is DeepSeek V3.1 open source?
Yes. Its weights are published.
When was DeepSeek V3.1 released?
August 2025.

Badge

DeepSeek V3.1 on Inference360: place 222 of 283, index 37.9
HTML
<a href="https://inference360.ai/models/deepseek-v3-1"><img src="https://inference360.ai/badges/deepseek-v3-1.svg" alt="DeepSeek V3.1 on Inference360: place 222 of 283, index 37.9" height="20"></a>
Markdown
[![DeepSeek V3.1 on Inference360: place 222 of 283, index 37.9](https://inference360.ai/badges/deepseek-v3-1.svg)](https://inference360.ai/models/deepseek-v3-1)

Compare with

Mistral Small 437.9North Mini Code38.0GPT-4.138.0Claude 3.7 Sonnet37.7DeepSeek V3.1 (Reasoning)43.0

More from DeepSeek

All models
DeepSeek V3.1 (Reasoning)43.0DeepSeek V3.1 TerminusUnrankedDeepSeek V3.2 ExpUnrankedDeepSeek-R1-052844.6DeepSeekMath V2Unranked