跳转到主要内容

DeepSeek-V2.5 (Dec '24)

DeepSeekDeepSeekOpen Weightdeepseek

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
164K
支持模态
text

能力雷达图

13
general
60
coding
76
reasoning
68
science估算
0
agents
0
multimodal

Science 在缺少专门科学评测时使用推理能力代理估算。

排行榜排名

领域#排名分数来源
通用能力榜471
14.0
AA
数学推理104
75.0
AA
推理能力49
69.0
LS

基准测试分数 (LLM Stats)

Code

HumanEval89.0%自报
Aider72.2%自报
SWE-Bench Verified16.8%自报

Communication

MT-Bench0.90 / 100自报

Creativity

AlignBench80.4%自报
Arena Hard76.2%自报
AlpacaEval 2.050.5%自报

Finance

MMLU80.4%自报

General

DS-FIM-Eval78.3%自报
LiveCodeBench(01-09)41.8%自报

Language

BBH84.3%自报

Math

GSM8k95.1%自报
MATH74.7%自报

Reasoning

HumanEval-Mul73.8%自报
DS-Arena-Code63.1%自报

AA 评测指数

Intelligence Index
12.5
Math 500
0.8

LLM Stats 分类评分

Communication
90
Roleplay
90
Finance
80
General
80
Healthcare
80
Language
80
Legal
80
Math
80
Writing
70
Creativity
70
Reasoning
70
Code
60
Frontend Development
20

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 tokens/s
首Token延迟0.00s
首回答延迟0.00s

可用提供商

(LS 内部计价单位)

暂无提供商数据

外部链接