modelgrep
D

DeepSeek: DeepSeek V3.1

deepseek/deepseek-chat-v3.1

84th smartest of 180ReasoningToolsJSON
Use via OpenRouter ↗
Intelligence
28.1
84th of 180
Design Elo
1167
Website
Speed
85
87th fastest
Latency
368ms
first token
Input price
$0.210
103rd cheapest
Context
164K
33K max out

How it compares

Smarter than53%
of all ranked models
Faster than72%
of all ranked models
Cheaper than65%
of all ranked models

Overview

DeepSeek-V3.1 is a large hybrid reasoning model (671B parameters, 37B active) that supports both thinking and non-thinking modes via prompt templates. It extends the DeepSeek-V3 base with a two-phase long-context...

Benchmarks

independent · via OpenRouter
Artificial Analysis50th percentile
Intelligence Index
28.1
Coding Index
28.4
Agentic Index
31.9
GPQA Diamond
74%
Humanity's Last Exam
6%
SciCode
37%
Tau²-Bench (agentic)
35%
Design Arena · Elo9,535 tournaments
Website
1167
codecategories
1164
3D
1158
Game Dev
1154
Data Viz
1143
UI Component
1140
svg
1023

Providers & pricing (7)

ProviderIn $/MOut $/MUptime
DeepInfrafp4$0.210$0.79097.2%
Novitafp8$0.270$1.00100%
SiliconFlowfp8$0.270$1.0097.4%
AtlasCloudfp8$0.300$0.950100%
WandBfp8$0.550$1.65100%
Google$0.600$1.7099.9%
SambaNovafp8$0.650$1.5098.9%

Specifications

Context window164K
Max output33K
Knowledge cutoffMar 2025
Input modalitiestext
Output modalitiestext
Prompt caching
Cache read price$0.130/M
ModeratedNo

DeepSeek V3.1 FAQ

How much does DeepSeek V3.1 cost?

DeepSeek V3.1 costs $0.210 per million input tokens and $0.790 per million output tokens via OpenRouter, making it 103rd cheapest of 298 paid models.

How smart is DeepSeek V3.1?

DeepSeek V3.1 scores 28.1 on the Artificial Analysis Intelligence Index, ranking 84th of 180 benchmarked models, with a GPQA Diamond score of 74%.

How fast is DeepSeek V3.1?

DeepSeek V3.1 generates around 85 tokens per second with 368ms time-to-first-token (p50), the 87th fastest tracked model.

What is DeepSeek V3.1's context window?

DeepSeek V3.1 supports a 164K-token context window and can output up to 33K tokens. It accepts text input.

Compare head-to-head