DeepSeek launched the V4-Flash stable version API for public beta yesterday, and today V4-Pro got its update too. The version number has changed to DeepSeek-V4-Pro-0813, with the main change being added support for the Responses API format.

This isn't a complicated matter on its own, but it's worth breaking down.

## What is the Responses API?

The Responses API is an interface format introduced by OpenAI in late 2024, combining the simplicity of Chat Completions with the state management capabilities of the Assistants API. In simple terms, it's the "next generation" of Chat Completions, better suited for agent scenarios — you don't need to manage conversation history or tool call states yourself; the server handles it for you.

What does DeepSeek adding this support to V4-Pro mean?

**Compatibility is productivity.** For development teams already using the Responses API, the cost of migrating to DeepSeek drops significantly. No need to change code architecture — just swap the base URL and API key and you're up and running.

## The positioning difference between Flash and Pro

The V4-Flash public beta from yesterday natively supports the Responses API with targeted adaptation for Codex, and its benchmark scores are quite impressive — Terminal Bench 2.1 at 82.7, Cybergym at 76.7, DSBench-FullStack at 68.7, and DSBench-Hard at 59.6.

Flash focuses on high cost-effectiveness and low latency, suited for large-scale calls; Pro leans more toward complex reasoning and high-quality output. With Pro now adding Responses API support, the product line's consistency is aligned.

One interesting detail: the Flash stable version's benchmark scores comprehensively surpass V4-Pro-Preview. This is normal — Flash is trained on a new architecture, while Pro-Preview is a preview version of the previous generation. But users can easily get confused: "Is Flash stronger than Pro?"

No. The stable V4-Pro (0813) should open up a wider gap — no matter how strong Flash is, it's still a lightweight contender.

## Observations

DeepSeek didn't make a big deal out of this update. No release announcement, no technical blog post — just a quiet version number change in the API docs.

But that's exactly why it signals they're busy with more important things. The V4 series product matrix is now clear: Flash handles speed and affordability, Pro handles power and stability. Combined with the adaptation for Codex and the Responses API, the goal is obvious — **pull developers from the OpenAI ecosystem over at the lowest possible cost.**

The current public beta is still in rapid iteration, and API behavior may be adjusted. Before integrating into production, it's recommended to run a round of validation in a test environment first.

You can find the version change log in the API docs. For differences in specific response formats, sending a few test requests and comparing them side by side is the most straightforward approach.