AuraTracer智迹闻
中文

EVENT DOSSIER

DeepSeek V4.1 Flash Beta: Uses a new structure and native multi-modal support

DeepSeek V4.1 Flash began internal testing on September 8, 2026. It supports multiple modalities natively and its billing system is the same as that of V4 Fla…

2026-09-08 08:00 Models 🔥 65.2 heat score baidu #16v2ex #10zhihu #12
2sources
1days unfolding
65.2heat score
3mentions
SummaryAI generated

On September 8, 2026, DeepSeek announced the beta testing service for the DeepSeek V4.1 Flash intermediate version. This model features a new structure and native multi-modal support, offering improvements in generation speed, cost efficiency, and Agent capabilities compared to the official version. Users can access the API without modifying their basic configurations; they only need to set the model name to deepseek-v4.1-flash-expires-on-0910. The current billing standards are the same as those for DeepSeek V4 Flash, with a limit of 20 concurrent requests per account. Previously, the DeepSeek-V4-Flash-Vision-Exp multi-modal visual understanding model was launched on August 21, supporting image description, text recognition, and chart analysis. The official feedback form asked users whether they believed this version could fully replace the online DeepSeek V4 Pro, indicating its aim to improve upon the Pro version. Practical tests show that V4.1 Fla…

Related eventsRELATED EVENTS
Quick factsQUICK FACTS
deepseek-v4.1-flash-expires-on-0910Model name
V4.1 FlashVersion type
Key entitiesKEY ENTITIES
Claude CodeDeepSeekPhoenix Network Technology

Event frameEVENT FRAME

Launch

AI · 中外媒体across 1 days

Status

DeepSeek V4.1 Flash began internal testing on September 8, 2026. It supports multiple modalities natively and its billing system is the same as that of V4 Fla…

Coverage · reports per dayLANGUAGE SPLIT

Entity relations
DeepSeek × Phoenix Netw…1Claude Code × DeepSeek1

Integrated timelineUNIFIED TIMELINE

  1. 2026-09-08

    DeepSeek V4.1 Flash began internal testing

    DeepSeek announced the start of internal testing for the intermediate version of DeepSeek V4.1 Flash, which features a new structure and native support for multiple modalities. Users can access the API without modifying their basic configurations; they only need to set the model name to deepseek-v4.1-flash-expires-on-09…

    2 reports

SignalsSIGNALS

Keyword heat
  • DeepSeek2
  • Phoenix Network Technology1
  • Claude Code1

All reports (2)SOURCES

凤凰网·科技 zh 2026-09-08 08:00

DeepSeek V4.1 Flash内测:采用新结构 原生多模态支持

9 月 8 日,DeepSeek 发布 DeepSeek V4.1 Flash 中间版本内测通知。该模型采用新结构,具备原生多模态支持,相比正式版能力更强、速度更快且成本更低。调用方式保持不变,需将模型名设置为 deepseek-v4.1-flash-expires-on-0910,当前计费与 deepseek-v4-flash 相同,每账号限流 20 并发。此前 DeepSeek-V4-Flash 正式版 API 于 7 月底上线公测,Agent 能力大幅增强;8 月 21 日,DeepSeek-V4-Flash-Vision-Exp 多模态视觉理解模型上线,支持输入 JPEG、PNG、GIF、WebP 等格式图片进行描述、截图文字识别及图表分析。

凤凰网·科技 zh 2026-09-08 08:00

DeepSeek新模型突然上线,Flash的模型,Pro的野心

DeepSeek 于近日开启 DeepSeek V4.1 Flash 限时内测,该版本原生支持多模态且生成速度更快、成本更低。用户接入 API 无需修改 base_url,仅需将模型名称设为 deepseek-v4.1-flash-expires-on-0910,测试期间计费标准与 V4 Flash 相同,单账号支持 20 路并发,该版本将于 9 月 10 日自动过期。官方在反馈问卷中询问用户是否认为该模型能全面替换线上 DeepSeek V4 Pro,暗示其旨在向 Pro 靠拢而非仅做提速降本。实测显示,V4.1 Flash 中文长文写作叙事节奏更自然,前端案例中科幻飞船驾驶舱 HUD 体现了 UI 与交互细节进步,但 macOS 网页操作系统制作表现中规中矩。综合来看,该版本生成速度快且稳定,若能在保持速…