https://www.deepseek.com/

DeepSeek Logo
Launching DeepSeek-V2.5, combining general and coding capabilities,
API and Web upgraded.
Brand new experience, redefining possibilities
 
Start Now
Free access to DeepSeek-V2.5.
Into the unknown.
 
Access API
128K context length.
$0.14-$0.28 for 1 million more.
Zhong Wen 
DeepSeek-V2.5 Capabilities

DeepSeek-V2.5 delivers impressive results on current major large
model leaderboards.

Places top 3 in AlignBench
Surpassing GPT-4 and close to GPT-4-Turbo
Ranks top-tier in MT-Bench
Rivaling LLaMA3-70B and outperforming Mixtral 8x22B
Specializes in math, code and reasoning
The open-source model and API support 128K context length

                  Open   Chinese   English  Knowledge Arithmetic Math Reasoning  Coding
                 source  General   General
                        AlignBench MT-Bench   MMLU      GSM8K    MATH    BBH    HumanEval
DeepSeek-V2.5    Yes    8.04       9.02     80.4      95.1       74.7 84.3      89.0
DeepSeek-V2      Yes    7.89       8.85     80.6      94.8       71.0 83.4      84.8
GPT-4-Turbo-1106 -      8.01       9.32     84.6      93.0       64.1 -         82.2
GPT-4-0613       -      7.53       8.96     86.4      92.0       52.9 83.1      84.1
GPT-3.5          -      6.08       8.21     70.0      57.1       34.1 66.6      48.1
Gemini1.5 Pro    -      7.33       8.93     81.9      91.7       58.5 84.0      71.9
Claude3 Opus     -      7.62       9.00     86.8      95.0       61.0 86.8      84.9
Claude3 Sonnet   -      6.70       8.47     79.0      92.3       40.5 82.9      73.0
Claude3 Haiku    -      6.42       8.39     75.2      88.9       40.9 73.7      75.9
abab-6.5         -      7.97       8.82     79.5      91.7       51.4 82.0      78.0
abab-6.5s        -      7.34       8.69     74.6      87.3       42.0 76.8      68.3
ERNIE-4.0        -      7.89       7.69     -         91.3       52.2 -         72.0
GLM-4            -      7.88       8.60     81.5      87.6       47.9 82.3      72.0
Moonshot-v1      -      7.22       8.59     -         89.5       44.2 -         82.9
Baichuan 3       -      -          8.70     81.7      88.2       49.2 84.5      70.1
Qwen1.5 72B      Yes    7.19       8.61     76.2      81.9       40.6 65.9      68.9
LLaMA 3 70B      Yes    7.42       8.95     80.3      93.2       48.5 80.1      76.2
Mixtral 8x22B    Yes    6.49       8.66     77.8      87.9       49.8 78.4      75.0

DeepSeek API Pricing
Per Million Input Tokens
0.14$
Per Million Output Tokens
0.28$
Why DeepSeek?
API Access
236B parameters
128K context (API)
Capable
$0.14/M input tokens
$0.28/M output tokens
Cost-effective
Compatible with
OpenAI API
Seamless
DeepSeek Logo
 
 
(c) 2024 DeepSeek. All rights reserved.
Zhe ICPBei 2023025841Hao Zhe Gong Wang An Bei  33010502011812 Hao 
Research
DeepSeek LLMDeepSeek CoderDeepSeek MathDeepSeek VLDeepSeek V2DeepSeek
Coder V2
Product
DeepSeek ChatDeepSeek PlatformAPI PricingService Status
Legal & Safety
Privacy PolicyTerms of Use