DeepSeek V4.1 Flash Officially Released: Native Multimodal, Outperforms V4 Pro with Lower API Prices

DeepSeek officially released the DeepSeek V4.1 Flash model on September 10, 2026. As the smallest version in the new model structure series, it features a 552B parameter Mixture of Experts (MoE) architecture and introduces a brand-new Causal-Encoder-Decoder structure. This asymmetric input-output design results in only 8B activated parameters for input and 16B for output, significantly reducing computational costs while maintaining high capability.

In terms of core capabilities, V4.1 Flash possesses native multimodal visual understanding, capable of processing image information without additional plugins. After training with larger-scale reinforcement learning, the model achieved excellent results in multiple benchmark tests, particularly on Agentic Benchmark, where its intelligence level was officially confirmed to comprehensively surpass flagship models including DeepSeek V4 Pro.

Addressing the common issue of high cache costs in Agent tasks, the new generation model has significantly optimized storage efficiency. Compared to the previous generation, V4.1 Flash reduces HBM requirements to 1/4 and SSD requirements to 1/8. Compared to the first-generation model, its KV Cache size is only 1/437 of the original, an innovation that substantially lowers usage costs in long-context scenarios.

Regarding service deployment and compatibility, V4.1 Flash is now live on the DeepSeek API. Users can invoke it simply by changing the model name to deepseek-flash. The legacy V4 Flash and V4 Flash Vision Exp models have been taken offline, but for compatibility reasons, requests accessing old names such as deepseek-v4-flash will be temporarily routed to V4.1 Flash.

Given the comprehensive advantages of V4.1 Flash in performance, cost, and speed, the official plan is to orderly phase out the V4 Pro model. Starting from 12:00 Beijing Time on September 14, 2026, until V4.1 Pro goes online, all requests to deepseek-v4-pro will be automatically routed to V4.1 Flash and billed at the new model's unit price.

Accompanying the model upgrade, DeepSeek has correspondingly lowered the API pricing for V4.1 Flash. To allocate resources more reasonably, the new pricing continues to adopt a peak-valley mechanism, with off-peak prices set at half of peak hours, aiming to encourage users to adjust task times according to actual needs. The new pricing system officially took effect at 12:00 on September 10, 2026.

In terms of ecosystem cooperation, Tencent's WorkBuddy, CodeBuddy, and OpenCode, as official partners, have fully integrated the DeepSeek V4.1 Flash model. Additionally, DeepSeek Harness version 0.1.5 has been updated to adapt to this model, adding file upload and sidebar preview features, further enhancing the developer experience.