DeepSeek did three things at once: moved its flagship product out of the testing phase, released its proprietary agent software as open source, and announced higher API prices.

The model: V4-Pro-0813

The deepseek-v4-pro endpoint now delivers build V4-Pro-0813. The model name, parameter count and one-million-token context window remain unchanged, and DeepSeek says existing integrations will keep running without any tweaks. In the app and on the web the model appears under "Expert Mode."

A new addition is native support for the OpenAI Responses API with Codex integration. Reasoning effort can be set to three levels — "low," "high" and "max" — with DeepSeek recommending the middle setting for everyday agent use.

Scores, in context

According to the company's own comparison table:

  • Terminal Bench 2.1: 72.1 → 87.9
  • DeepSWE: 12.8 → 62.7
  • On several agent benchmarks the model beats Claude Opus 4.8

Artificial Analysis backs up the improvement but also places it: V4-Pro climbs from 45 to 53 on the Intelligence Index, tying GLM-5.2. That is still behind Muse Spark at 57, Qwen 3.8 Max at 58 and Kimi K3 at 60. Claude Opus 5 sits at the top with 63.

DeepSeek has not published weights for the new build; the April preview version is still what sits on Hugging Face.

Pressure from its own small model

Part of the reason for the update comes from the company's own line-up. The smaller V4 Flash had been closing in on the flagship: with update 0731 at the end of July, Flash practically matched Pro Preview on the Intelligence Index while costing a fraction of the price.

It is a problem model makers meet often: your own cheap model erodes the reason your expensive one exists.

Time-based pricing

The pricing change is the unusual part. API prices rise overall, but new time-based rates arrive with them: usage outside Chinese business hours becomes cheaper. Repeated data retrievals get pricier.

Time-based pricing is a familiar method in cloud services, nudging load toward off-peak hours. It is not yet common in AI APIs, and it offers a hint about how GPU capacity fluctuates over the course of a day.

What the three moves mean together

Announcing all three on the same day is not incidental. Open-sourcing the agent software makes the model usable without tying the user to DeepSeek's infrastructure — a distribution tool rather than a lock-in one.

The price rise points the other way: it is the move of a company facing demand it cannot meet or costs that have grown. The addition of time-based rates supports that reading; the constraint looks less like total capacity than its distribution across the day.

That the weights have not yet been published is its own signal. DeepSeek is known for its open-weight tradition; holding the new build back suggests the balance between its commercial and open sides may be shifting.