DeepSeek released DeepSeek V4 Flash in public beta, keeping the existing API base while adding a new deepseek-v4-flash model name.
What changed
DeepSeek reports substantial gains across several coding and agent benchmarks. The model also natively supports the Responses API format and is specifically adapted for coding-agent workflows.
Why it matters
The unchanged calling pattern makes trials relatively straightforward, while Responses API compatibility gives developers another way to connect the model to tool-using systems. The update applies to the API model; DeepSeek says its app, web product, and V4 Pro API are unchanged.
Our take
Benchmark claims are a starting point, not a deployment decision. Test multi-file edits, tool-call recovery, instruction persistence, and total cost on your own repositories. Read our DeepSeek vs. ChatGPT comparison and Chinese AI models for developers.