HomeDeepSeek-V2-Chat-0628
236B
Model parameter quantity
DeepSeek
Affiliated organization
Open Source
License Type
May 6, 2024
Release time
Related figures
Model Introduction
DeepSeek-V2 is a strong Mixture-of-Experts (MoE) language model characterized by economical training and efficient inference. It comprises 236B total parameters, of which 21B are activated for each token. Compared with DeepSeek 67B, DeepSeek-V2 achieves stronger performance, and meanwhile saves 42.5% of training costs, reduces the KV cache by 93.3%, and boosts the maximum generation throughput to 5.76 times.
Comprehensive score
Language dialogue
Knowledge reserve
Reasoning association
Mathematical calculation
Code writing
Command following
Swipe left and right to view more
Language comprehension ability
Often makes semantic misjudgments, leading to obvious logical disconnects in responses.
4.6
Knowledge coverage scope
Possesses core knowledge of mainstream disciplines, but has limited coverage of cutting-edge interdisciplinary fields.
7.8
Reasoning ability
Unable to maintain coherent reasoning chains, often causing inverted causality or miscalculations.
4.7
Model comparison
DeepSeek-V2-Chat-0628 vs Qwen2.5-7B-Instruct
Like Qwen2, the Qwen2.5 language models support up to 128K tokens and can generate up to 8K tokens. They also maintain multilingual support for over 29 languages, including Chinese, English, French, Spanish, Portuguese, German, Italian, Russian, Japanese, Korean, Vietnamese, Thai, Arabic, and more.
DeepSeek-V2-Chat-0628 vs Hunyuan-T1-20250822
The deep reasoning model independently developed by Tencent adopts the version number hunyuan-t1-20250822.
DeepSeek-V2-Chat-0628 vs Spark-X1
The inference model Spark X1 released by iFlytek, on the basis of leading domestic mathematical tasks, benchmarks the performance of general tasks such as inference, text generation, and language understanding against OpenAI o1 and DeepSeek R1.
Related model
DeepSeek-V3.2
The latest version of Deepseek V3 series models.
DeepSeek-V3.2-Exp
The latest experimental version of Deepseek V3 series models.
DeepSeek-R1-0528
The latest version of Deepseek R1.
DeepSeek-V3-0324
DeepSeek-V3 outperforms other open-source models such as Qwen2.5-72B and Llama-3.1-405B in multiple evaluations and matches the performance of top-tier closed-source models like GPT-4 and Claude-3.5-Sonnet.
DeepSeek-R1-0528
The latest version of Deepseek R1.
Relevant documents
China Telecom Invests in Mianbi Intelligence, Raises Capital to 713,000 Yuan for LLM & Data Infra
The "national team" and the leading figure from Tsinghua University in the large model space are deepening their strategic alignment. On March 1, 2026, according to the latest business registration data from Qichacha, Beijing Mianbi Intelligent Techn
Taotian Group Accelerates AI-Native Restructuring, Grants Interns Free Token Quotas
TaoTian Group recently introduced the "AI Productivity Plan," designed to accelerate the integration of AI technology into e-commerce operations and R&D workflows through resource allocation and tool subsidies. The program is now available to all int
Glean targets enterprise AI infrastructure in land grab
The race to dominate enterprise AI is accelerating. Microsoft is embedding Copilot into Office, Google is integrating Gemini into Workspace, and both OpenAI and Anthropic are selling directly to corporations. Meanwhile, nearly every SaaS vendor now i
Tibetan Language Model Earns Acclaim at Beijing Science and Technology Fair
At the 28th China Beijing International Science and Technology Industry Expo, a large language model that can "understand" Tibetan and generate content using Tibetan-style thinking drew significant attention. Developed by a research team from Tibet U
Anthropic Merges with Google in $20B Bet, AI Computing Competition Intensifies
As competition in artificial intelligence intensifies, leading AI startups like Anthropic have made a major move. Sources say the company has officially committed to paying Google up to $20 billion over the next five years. This large sum will mainly
