DeepSeek-V4 API Receives Update and is Available Today, Expanding 1M Context Length Access
Two new model variants rival top closed-source models on agentic coding.
DeepSeek has released the V4 Preview, open-sourcing two powerful models: DeepSeek-V4-Pro with 1.6 trillion total parameters (49B active) and DeepSeek-V4-Flash with 284B total (13B active). Both achieve state-of-the-art results among open models in agentic coding, world knowledge, and reasoning, rivaling top closed-source models like Gemini-3.1-Pro. The models support a 1M context length by default, enabled by novel attention mechanisms including token-wise compression and DeepSeek Sparse Attention (DSA), which drastically reduce compute and memory costs.
The models are designed for agentic capabilities, seamlessly integrating with tools like Claude Code, OpenClaw, and OpenCode. The API is updated today, supporting OpenAI and Anthropic endpoints; users can switch to model names 'deepseek-v4-pro' or 'deepseek-v4-flash'. The older 'deepseek-chat' and 'deepseek-reasoner' will be retired after July 24, 2026. This release positions DeepSeek as a leader in cost-effective, high-performance open-source AI with long context windows and strong agent performance.
- DeepSeek-V4-Pro: 1.6T total / 49B active params, rivals top closed models in reasoning and agentic coding.
- DeepSeek-V4-Flash: 284B total / 13B active params, offers fast, economical performance with near-Pro reasoning.
- Both models support 1M context via token-wise compression and DeepSeek Sparse Attention; API updated today with OpenAI/Anthropic compatibility.
Why It Matters
DeepSeek's open-source models deliver 1M context and agentic coding at scale, democratizing access to cutting-edge AI.