DataLearner logoDataLearnerAI
Latest AI Insights
Model Leaderboards
Benchmarks
Model Directory
Model Comparison
Resource Center
Tools
LanguageEnglish
DataLearner logoDataLearner AI

A knowledge platform focused on LLM benchmarking, datasets, and practical instruction with continuously updated capability maps.

Products

  • Leaderboards
  • Model comparison
  • Datasets

Resources

  • Tutorials
  • Editorial
  • Tool directory

Company

  • About
  • Privacy policy
  • Data methodology
  • Contact

© 2026 DataLearner AI. DataLearner curates industry data and case studies so researchers, enterprises, and developers can rely on trustworthy intelligence.

Privacy policyTerms of service
Page navigation
目录
Model catalogDeepSeek-V4-Flash
DE

DeepSeek-V4-Flash

推理大模型

DeepSeek V4 Flash

Release date: 2026-04-24更新于: 2026-04-24 13:39:57.266知识截止: 2025-05257
Live demoGitHubHugging FaceCompare
Parameters
2840.0亿
Context length
1M
Chinese support
Supported
Reasoning ability

DeepSeek V4 Flash is an AI model published by DeepSeek-AI, released on 2026-04-24, for 推理大模型, with 2840.0B parameters, and 1M tokens context length, under the MIT License license.

Data sourced primarily from official releases (GitHub, Hugging Face, papers), then benchmark leaderboards, then third-party evaluators. Learn about our data methodology

DeepSeek-V4-Flash

Model basics

Reasoning traces
Supported
Thinking modes
Thinking Level · Max (Default)Standard ModeThinking Level · High
Context length
1M tokens
Max output length
384000 tokens
Model type
推理大模型
Release date
2026-04-24
Model file size
No data
MoE architecture
Yes
Total params / Active params
2840.0B / 130B
Knowledge cutoff
2025-05
DeepSeek-V4-Flash

Open source & experience

Code license
MIT License
Weights license
MIT License- 免费商用授权
GitHub repo
GitHub link unavailable
Hugging Face
https://huggingface.co/collections/deepseek-ai/deepseek-v4
Live demo
https://chat.deepseek.com
DeepSeek-V4-Flash

Official resources

Paper
DeepSeek-V4 Technical Report
DataLearnerAI blog
No blog post yet
DeepSeek-V4-Flash

API details

API speed
4/5
💡Default unit: $/1M tokens. If vendors use other units, follow their published pricing.
Learn about pricing modes
Standard
TypeConditionInputOutput
Text-$0.140/ 1M$0.280/ 1M
Cache PricingPrompt Cache
TypeTTLWriteRead
Text1d$0.140/ 1M$0.028/ 1M
DeepSeek-V4-Flash

Benchmark Results

DeepSeek-V4-Flash currently shows benchmark results led by LiveCodeBench (4 / 118, score 91.60), MMLU Pro (13 / 124, score 86.40), IMO-AnswerBench (2 / 17, score 88.40). This page also consolidates core specs, context limits, and API pricing so you can evaluate the model from benchmark results and deployment constraints together.

Thinking
All modesNormalThinking
Thinking mode details (2)
All thinking modesDefault (Max)Deep
Tool usage
All modesWith toolsNo tools
Internet
All modesOfflineInternet enabled

编程与软件工程

6 evaluations
Benchmark / mode
Score
Rank/total
SWE-bench Verified
HighTools
78.60
17 / 103
SWE-bench Verified
DeepTools
79
15 / 103
SWE-bench Multilingual
HighTools
70.20
11 / 17
SWE-bench Multilingual
DeepTools
73.30
7 / 17
SWE-Bench Pro - Public
HighTools
52.30
21 / 36
SWE-Bench Pro - Public
DeepTools
52.60
19 / 36

AI Agent - 工具使用

2 evaluations
Benchmark / mode
Score
Rank/total
Terminal Bench 2.0
HighTools
56.60
24 / 43
Terminal Bench 2.0
DeepTools
56.90
22 / 43

生产力知识

1 evaluations
Benchmark / mode
Score
Rank/total
GDPval-AA
DeepTools
1395
4 / 20
View benchmark analysisCompare with other models
DeepSeek-V4-Flash

Publisher

DeepSeek-AI
DeepSeek-AI
View publisher details
DeepSeek V4 Flash

Model Overview

DeepSeek-V4-Flash 预览版:普惠经济的百万上下文大模型

DeepSeek-V4-Flash 是 DeepSeek 于 2026 年 4 月 24 日正式发布并开源的旗舰级大语言模型预览版,属于 DeepSeek-V4 系列的高性价比型号。该模型旨在提供远低于行业平均水平的推理成本,同时保持百万级上下文、Agent 能力和顶级推理性能,适合需要高吞吐、低成本的生产级应用。

架构与技术规格

DeepSeek-V4-Flash 采用了与 Pro 版本一脉相承的混合专家(MoE)架构,总参数量为 2840 亿(284B),每次推理激活参数约 130 亿(13B)。其上下文窗口原生支持 100 万 token(1M),最大输出长度可达 384K token。V4 系列均引入了创新的混合注意力机制,融合了压缩稀疏注意力(CSA)与重压缩注意力(HCA),并结合 DSA 稀疏注意力,这使得处理百万级上下文时的计算和显存需求大幅降低。其中,V4-Flash 预训练数据量高达 32 万亿 token,并同样采用了 Muon 优化器、流形约束超连接(mHC)等新型训练策略。

核心能力与支持模态

DeepSeek-V4-Flash 目前为纯文本模型,不支持视觉输入或图像识别等任务。其核心能力聚焦于 Agent 能力和推理性能:V4-Flash 在提供足够思考预算后,其推理能力可极为接近 Pro 版本,但在纯知识性问答和复杂 Agent 任务上受限于参数量,表现略逊于 Pro 版。此外,模型同时支持非思考模式与思考模式,用户可通过 reasoning_effort 参数调节思考强度以应对复杂推理任务。

性能评价

根据官方公布的数据,DeepSeek-V4-Flash 在提供足够推理预算的前提下,其推理能力接近 V4-Pro,但在纯知识和复杂 Agent 任务上受到参数量的限制。官方已将其作为 deepseek-chat 和 deepseek-reasoner 的替代品,这两个旧名称将于 2026 年 7 月 24 日停用。

应用场景与限制

官方推荐场景包括:通用对话、基础文本生成、高并发和低成本的线上服务、Agent 工作流中的低延迟任务。当前局限在于:尚不支持图像等多模态输入;在纯知识和极其复杂的 Agent 任务上性能不如 V4-Pro 等更大的模型;作为预览版,未来 API 的稳定性和功能可能存在调整。

访问方式与许可

模型已全面开源,权重和技术报告可通过 Hugging Face 和魔搭社区获取。许可证采用 MIT License,允许商用、修改和再分发。API 服务已同步上线,开发者通过修改 model_name 为 deepseek-v4-flash 即可调用,兼容 OpenAI ChatCompletions 和 Anthropic 接口格式。此外,旧的 deepseek-chat 和 deepseek-reasoner 名称将分别指向 V4-Flash 的非思考模式和思考模式,直至 2026 年 7 月 24 日弃用。

DataLearner 官方微信

欢迎关注 DataLearner 官方微信,获得最新 AI 技术推送

DataLearner 官方微信二维码