DeepSeek-V2.5 (Dec '24)
DeepSeekDeepSeek开源权重deepseek
描述
DeepSeek-V2.5 is an upgraded version that combines DeepSeek-V2-Chat and DeepSeek-Coder-V2-Instruct, integrating general and coding abilities. It better aligns with human preferences and has been optimized in various aspects, including writing and instruction following.
发布日期
2024-12-10
参数规模
236.0B
上下文长度
—
支持模态
text
能力雷达图
7
general
60
coding
76
reasoning
68
science估算
71
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 |
|---|---|---|---|
| 通用能力榜 | 571 | 9.0 | AA |
基准测试分数 (LLM Stats)
(LLM Stats (zeroeval))Code
HumanEvalOpenAI (2021)
89.0%自报
Aider
72.2%自报
SWE-Bench Verified
16.8%自报
Communication
MT-Bench
0.90 / 100自报
Creativity
AlignBench
80.4%自报
Arena Hard
76.2%自报
AlpacaEval 2.0
50.5%自报
Finance
MMLU
80.4%自报
General
DS-FIM-Eval
78.3%自报
LiveCodeBench(01-09)
41.8%自报
Language
BBH
84.3%自报
Math
GSM8k
95.1%自报
MATH
74.7%自报
Reasoning
HumanEval-Mul
73.8%自报
DS-Arena-Code
63.1%自报
AA 评测指数
(Artificial Analysis)Intelligence Index(Artificial Analysis)6.5
Math 500(OpenAI (2024), subset of Hendrycks et al. MATH (2021))0.8
LLM Stats 分类评分
(LLM Stats (zeroeval))Roleplay90
Communication90
Legal80
Math80
Language80
Finance80
Healthcare80
Reasoning70
General70
Creativity70
Writing70
Code60
Frontend Development20
定价
输入价格免费
输出价格免费
混合价格(3:1)免费
速度
Tokens/秒0.0
首Token延迟0.00s
首回答延迟0.00s
供应商价格排行
暂无提供商数据