DeepSeek V4.1 Flash has been upgraded to DeepSeek-V4.1-Flash-1017, delivering substantial benchmark improvements across coding, reasoning, and multimodal tasks. The update also reduces task runtime and output length, achieving a new performance-cost Pareto optimum. Additionally, DeepSeek's in-house search API has been opened up, supporting hundreds of billions of indexed web pages with multilingual search.

DeepSeek-V4.1-Flash is the smallest model in a new architecture family featuring native multimodal visual understanding, designed for higher capability ceiling, faster inference, and higher throughput. It outperforms the previous DeepSeek V4 Pro across performance, cost, speed, and total time, leading to the planned retirement of V4 Pro. API prices have been reduced with this release.

DeepSeek released a new experimental multimodal vision understanding model called DeepSeek-V4-Flash-Vision-Exp, accessible via API with model='deepseek-v4-flash-vision-exp'. It maintains the same pure-text capabilities as DeepSeek-V4-Flash while delivering significant improvements on agent benchmarks requiring visual understanding. Its multimodal agent capabilities are described as close to Opus-4.8.

The GA (General Availability) release of DeepSeek-V4-Pro has been rolled out on APP, Web, and API with significantly enhanced agent capabilities showing major performance improvements in production environments. The release adds native support for the OpenAI Responses API format with Codex integration, and introduces more flexible thinking effort control with three levels (low/high/max) for both V4-Pro and V4-Flash. API pricing is also being updated with peak/off-peak pricing structure effective August 16, 2026.

DeepSeek-V4-Flash API is now in official public beta release, with significantly enhanced agent capabilities far exceeding V4-Pro-Preview across multiple benchmarks. The model natively supports the Responses API format and is specifically adapted for Codex. It keeps the same model architecture and size as DeepSeek-V4-Flash-Preview but was re-post-trained.

DeepSeek-V4-Flash has been officially released into public beta via the API, accessible by setting the model name to 'deepseek-v4-flash'. It features significantly enhanced agent capabilities with benchmark results far exceeding V4-Pro-Preview across multiple agent benchmarks. The model keeps the same architecture and size as DeepSeek-V4-Flash-Preview but was re-post-trained, and now natively supports the Responses API format with specific adaptation for Codex.

DeepSeek API now supports V4-Pro and V4-Flash models, available via both OpenAI ChatCompletions and Anthropic interfaces. The legacy model names 'deepseek-chat' and 'deepseek-reasoner' will be discontinued on 2026-07-24, currently pointing to non-thinking and thinking modes of deepseek-v4-flash respectively. Models are accessible via model parameters 'deepseek-v4-pro' or 'deepseek-v4-flash'.