Skip to main content

DeepSeek-V2.5 (Dec '24)

DeepSeekDeepSeekOpen Weightdeepseek

Description

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.

Release Date
2024-12-10
Parameters
236.0B
Context Length
164K
Modalities
text

Capability Radar

13
general
60
coding
76
reasoning
68
scienceest.
0
agents
0
multimodal

Science uses a reasoning proxy when dedicated science benchmarks are unavailable.

Rankings

Domain#RankScoreSource
General Ranking471
14.0
AA
Math Reasoning104
75.0
AA
Reasoning49
69.0
LS

Benchmark Scores (LLM Stats)

Code

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

Communication

MT-Bench0.90 / 100SR

Creativity

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

Finance

MMLU80.4%SR

General

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

Language

BBH84.3%SR

Math

GSM8k95.1%SR
MATH74.7%SR

Reasoning

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

AA Evaluation Indices

Intelligence Index
12.5
Math 500
0.8

LLM Stats Category Scores

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

Pricing

Input PriceFree
Output PriceFree
Blended Price (3:1)Free

Speed

Tokens/sec0.0 tokens/s
Time to First Token0.00s
Time to Answer0.00s

Available Providers

(LS internal units)

No provider data available

External Sources