DeepSeek: DeepSeek V4 Flash Vision Exp下线

DeepSeek: DeepSeek V4.1 Flash 1017

DeepSeek V4.1 Flash is a 552B-parameter MoE model built on a new causal encoder-decoder architecture, designed for higher capability, faster reasoning, higher throughput, and lower serving cost. It natively supports multimodal visual understanding and delivers flagship-level intelligence with significantly reduced KV cache requirements, making it well suited for high-throughput and cost-sensitive agentic workloads.

DeepSeek: DeepSeek V4 Pro 0813

**In addition to DeepSeek, TencentCloud and AlibabaCloud now also offer the official 0731 version, while other providers are still using the old 0424 version**. . DeepSeek V4 Pro is DeepSeek’s flagship open-weight Mixture-of-Experts model, released in preview on April 24, 2026. It features 1.6 trillion total parameters, with 49 billion activated per token, and supports a one-million-token context window. Designed for advanced reasoning, coding, STEM tasks, and autonomous agents, it delivers state-of-the-art performance among open models and rivals leading proprietary systems. The model supports both thinking and non-thinking modes and is available through DeepSeek’s applications and APIs.

DeepSeek: DeepSeek V4 Flash 0731

DeepSeek-V4-Flash is the efficiency-oriented variant of the DeepSeek V4 series, released as a preview and open-sourced alongside the flagship V4-Pro. It is designed for developers who need the V4 generation's long-context and reasoning capability at a faster, more economical API tier. Compared with V4-Pro, V4-Flash uses smaller total parameters and active parameters, resulting in faster response times and lower API cost. It retains reasoning capability close to V4-Pro and matches V4-Pro on simple agent tasks, with a measurable gap appearing only on the most demanding agent workflows. World knowledge is slightly below V4-Pro but remains competitive within the open-source landscape. Like V4-Pro, V4-Flash inherits the new attention mechanism built on token-dimension compression and DeepSeek Sparse Attention (DSA), supports a 1M context window as standard, and offers both thinking and non-thinking modes with a reasoning_effort parameter (high / max). Note on migration: the legacy model names deepseek-chat and deepseek-reasoner currently route to V4-Flash in non-thinking and thinking mode respectively, and will be retired on 2026-07-24. Existing integrations should migrate to the explicit model names deepseek-v4-flash or deepseek-v4-pro. The model is accessible via OpenAI ChatCompletions and Anthropic interfaces, with weights open-sourced on Hugging Face and ModelScope.