DeepSeek Plans to Release V4.1 Flash Around September 10

en.Wedoany.com Reported - On September 9, DeepSeek's open platform issued a notice stating that the V4.1 Flash model is scheduled to officially launch around September 10, Beijing time. DeepSeek stated that after extensive internal and external testing, V4.1 Flash has comprehensively surpassed the existing V4 Pro in metrics such as performance, cost, speed, and total task completion time. After V4.1 Flash officially launches and before V4.1 Pro goes live, all requests originally routed to V4 Pro will be redirected to V4.1 Flash and billed at V4.1 Flash pricing.

The V4.1 Flash release represents a further iteration on the DeepSeek V4 series. DeepSeek released the V4 preview version in April this year, establishing two model tiers—V4 Pro and V4 Flash—where V4 Flash adopts a smaller parameter scale, emphasizing faster response speed and lower API costs, while supporting a 1 million Token context window and both Thinking and Non-Thinking modes alongside V4 Pro.

Prior to the official release, DeepSeek opened testing for the V4.1 Flash intermediate version on September 8. The test version adopts a new model architecture and supports native multimodal input. The model is invoked under the name deepseek-v4.1-flash-expires-on-0910. During the testing phase, V4 Flash billing standards remain in effect, with a concurrency limit of 20 per account. In its testing feedback, DeepSeek also directly asked users whether the V4.1 Flash intermediate version could fully replace the production V4 Pro.

The pricing structure will also be adjusted accordingly. Starting at 12:00 PM Beijing time on September 10, DeepSeek will lower Flash series API prices: during off-peak hours, the cache-hit input price will be adjusted to RMB 0.02 per million tokens, cache-miss input to RMB 1, and output to RMB 4; peak-hour prices will be double those of off-peak hours. Compared with the previous pricing, the cache-hit input price drops from RMB 0.05 to RMB 0.02, a 60% reduction; cache-miss input drops from RMB 1.5 to RMB 1, and output drops from RMB 4.5 to RMB 4.

The DeepSeek V4 series has now entered a phase of continuous expansion in Agent and multimodal capabilities. At the end of July, the V4 Flash production API went live, followed by the V4 Flash Vision experimental version on August 21, adding image input and visual understanding capabilities. The V4.1 Flash test version further introduces a new model architecture and native multimodal support. The official launch date for V4.1 Pro has not yet been announced.