DeepSeek-V4-Pro reaches general availability

Background

DeepSeek-V4 was released in preview on 24 April 2026, with two models, V4-Pro and V4-Flash ^[3]. V4-Flash became official on 31 July 2026, after both models had first appeared as previews on 24 April 2026 ^[2][3]. The general-availability release of V4-Pro followed on 13 August 2026, moving that model from preview to a production version ^[1].

What happened

On 13 August 2026 DeepSeek launched the general-availability version of DeepSeek-V4-Pro, citing major agent upgrades that DeepSeek said brought strong production gains ^[1]. The API model names were unchanged ^[1].

Key details

Reasoning effort became adjustable for both V4-Pro and V4-Flash. DeepSeek recommended a low setting for simple tasks, a high setting for daily agent workflows and a max setting for complex tasks ^[1].

The GA release added native support for the OpenAI Responses API, with one-click setup for Codex ^[1]. Codex is OpenAI's coding agent, which users can connect to the model through this setup ^[1].

V4-Pro also became available in the DeepSeek app and on the web through 'Expert Mode' ^[1].

Pricing changed as well. DeepSeek introduced peak and off-peak API rates, with off-peak prices 50 percent lower. The off-peak rate took effect on 16 August 2026 at 16:00 UTC ^[1].

Impact

The GA release brought V4-Pro to production status alongside the other models in the family. It also gave developers more control over how much reasoning each request uses and, through the peak and off-peak rates, when a request is billed ^[1].

Terms used in the release

General availability, usually shortened to GA, means that a product is offered to all users in its production form after any preview period has ended. A model that has reached GA is one that developers can build live services on.

Reasoning effort is a setting that controls how much reasoning a model carries out before it answers. A low setting gives faster and cheaper responses, while a high setting spends more computation on hard problems. The OpenAI Responses API is one of the interfaces through which applications send requests to a model, and DeepSeek's support for it is part of the same release ^[1].

Peak and off-peak pricing means that the same model is charged at different rates depending on the time at which a request is sent. Providers use such schedules to reflect changing demand on their servers. DeepSeek's off-peak rate is 50 percent lower than the standard rate and took effect at 16:00 UTC on 16 August 2026 ^[1].