DeepSeek V3.2 (Non-reasoning)
DeepSeekDeepSeek开源权重MIT · 商用许可
描述
DeepSeek-V3.2 is a 685B-parameter MoE model that harmonizes high computational efficiency with superior reasoning and agent performance. It introduces DeepSeek Sparse Attention (DSA) for efficient long-context processing, a scalable reinforcement learning post-training framework, and large-scale agentic task synthesis covering 1,800+ environments. V3.2 achieves GPT-5-level performance across reasoning, coding, and agentic benchmarks, with gold-medal results from its Speciale variant on IMO, IOI, ICPC World Finals, and CMO 2025.
发布日期
2025-12-01
参数规模
685.0B
上下文长度
164K
支持模态
text
能力雷达图
70
general
70
coding
80
reasoning
68
science估算
50
agents
0
multimodal
缺少专门科学评测时,Science 由 LLM Stats 科学得分或推理能力估算。
排行榜排名
| 领域 | #排名 | 分数 | 来源 |
|---|---|---|---|
| 智能体能力模型榜 | 128 | 36.0 | LS |
基准测试分数 (LLM Stats)
(LLM Stats (zeroeval))Agents
t2-bench
80.3%自报
MCP-Universe
45.9%自报
MCP-Mark
38.0%自报
Toolathlon
35.2%自报
Factuality
SimpleQA
97.1%自报
General
Aider-Polyglot
74.5%自报
Language
MMLU-Pro
85.0%自报
Math
AIME 2025
89.3%自报
HMMT 2025
83.6%自报
IMO-AnswerBench
78.3%自报
CodeForces
0.71 / 3000自报
Reasoning
GPQANYU + Cohere + Anthropic (2023)
79.9%自报
LiveCodeBench
74.1%自报
SWE-Bench Verified
67.8%自报
SWE-bench Multilingual
57.9%自报
BrowseComp-zh
47.9%自报
Terminal-Bench 2.0Stanford × Laude Institute (2026)
46.4%自报
BrowseCompOpenAI (2025)
40.1%自报
Terminal-Bench
37.7%自报
Humanity's Last Exam
19.8%自报
AA 评测指数
(Artificial Analysis)暂无 AA 评测数据
LLM Stats 分类评分
(LLM Stats (zeroeval))Language80
Legal80
Math80
Physics80
Finance80
Healthcare80
Biology80
Chemistry80
Reasoning70
Frontend Development70
General70
Code70
Search60
Agents50
Tool Calling50
Vision40
定价
输入价格$0.28 / 1M tokens
输出价格$0.42 / 1M tokens
混合价格(3:1)$0.29 / 1M tokens
缓存读取价格$0.13 / 1M tokens
速度
暂无速度数据
供应商价格排行
供应商价格排行
2 个供应商
最便宜: DeepInfra最贵: DeepSeek
供应商输入输出
1DeepInfra最便宜
$0.26
$0.38
2DeepSeek主要
$0.28
$0.42
比较该模型在不同 API 供应商之间的定价。