EVENT DOSSIER
On September 8, 2026, DeepSeek announced the start of beta testing for the V4.1 Flash intermediate version. This version features a new model structure, supports native multimodality, and offers faster processing times and lower costs. It performed nearly on par with Opus-4.8 in the multimodal AgentBenchmark. Test feedback showed that its end-to-end speed was significantly better than the previous version: SVG code generation was 6 times faster, and long context retrieval was more than 5 times faster. The beta testing period will end automatically on September 10. However, on September 9, DeepSeek announced that it would adjust the pricing of the Flash series models starting the next day (September 10), with a maximum discount of 60%. After this adjustment, the unit price for cache hits during idle periods dropped to 0.02 yuan, for missed hits to 1 yuan, and for output operations to 4 yuan; prices during peak hours doubled accordingly. Compared to the previous pricing levels, comprehensive estimates indicate that developer costs may decrease by about 40%, and some prices have already…
DeepSeek DeepSeek V4.1-Flash 开启内测，原生多模态支持
On September 8th, DeepSeek announced that the V4.1 Flash intermediate version was starting beta testing. This model features a new structure and supports native multi-modal processing, making it faster and less expensive. Developers need to call the model with the name deepseek-v4.1-flash-expires-on-0910.
3 reportsOn September 9th, DeepSeek announced that starting at 12:00 on September 10, 2026, Beijing time, the pricing of the flash series models would be adjusted, with up to 60% reductions. After this adjustment, the unit price for cache hits during idle periods will be 0.02 yuan, and for misses, it will be 1 yuan. The output unit price will be…
2 reports9 月 8 日，DeepSeek 宣布开启 V4.1-Flash 中间版本内测。该版本采用新模型结构，具备原生多模态支持、能力更强、速度更快及成本更低的特点。用户可保持 base_url 不变，将模型名设置为 deepseek-v4.1-flash-expires-on-0910 进行调用，当前计费与 deepseek-v4-flash 相同。此前 DeepSeek 已推出 V4-Flash-Vision-Exp 多模态视觉理解模型，在纯文本能力上与正式版持平，在多模态 AgentBenchmark 上实现大幅跃升，接近 Opus-4.8 水平。此外，DeepSeek Harness v0.1 开发者预览版已于 8 月 13 日全球开放测试并公开源代码，不到 24 小时在 GitHub 获超 7 万颗星。
9 月 8 日，DeepSeek 宣布 V4.1 Flash 中间版本开启内测。该模型采用新结构并支持原生多模态，速度更快、成本更低。开发者需将模型名设为 deepseek-v4.1-flash-expires-on-0910 调用，每账号限流 20 并发，测试期仅至 9 月 10 日。首批反馈显示，V4.1 Flash 端到端速度全面碾压旧版，SVG 代码生成快 6 倍、长上下文检索快 5 倍多。官方称计费与 V4 Flash 相同，但认为底层推理效率提升降低了实际成本；开发者反映账单未降，推测因速度加快导致单位时间 Token 消耗增加。DeepSeek 近两个月密集更新模型及 Agent 工具，旨在补齐产品体系，但据 Artificial Analysis 数据，其智能指数 36 分仍居第 15 位。
9 月 8 日下午，DeepSeek 发布 V4.1 Flash 中间测试版本开启内测，该版本非正式发布，仅供开发者提前试用且测试窗口为期两天。此版本名称 deepseek-v4.1-flash-expires-on-0910 表明其将于 9 月 10 日自动下线。
9 月 9 日，DeepSeek 宣布将于北京时间 2026 年 9 月 10 日 12:00 起调整 flash 系列模型定价，最高降幅达 60%。此次调价后，空闲时段缓存命中输入单价降至 0.02 元、未命中输入为 1 元，输出单价为 4 元；高峰时段价格相应翻倍。相比此前涨价后的水平，缓存命中输入降幅约 60%，未命中输入降幅约 33%，输出降幅约 11%。综合估算显示开发者调用成本可能下降约 40%，实际降幅受缓存命中率等因素影响。DeepSeek 仍维持峰谷分时定价机制，工作日特定时段为高峰。此次降价距其 8 月 V4 Pro 系列大幅涨价不足一个月，部分价格已回调至涨价前水平，但输出价格仍高于此前标准。背景显示 DeepSeek 正测试新一代 v4.1 flash 模型内测版，该版本主打速度更快、…
DeepSeek V4.1 Flash 在测试中完成 14 组任务共消耗 3 亿 token，其费用高于上一代产品。此次实测对比显示，该版本在相同任务量下的 Token 成本出现上涨。