DeepSeek puts V4 Pro into GA and switches the API to peak pricing

DeepSeek’s API changelog for August 13 said the GA release of DeepSeek-V4-Pro had rolled out on the app, web, and API. The dated checkpoint DeepSeek-V4-Pro-0813 also appeared on Hugging Face the same day under an MIT license. A preview had shipped in April.

What the company published

The changelog listed stronger agent scores, native support for the OpenAI Responses API format (adapted for Codex), and a three-rung thinking-effort ladder — low, high, max — on both V4-Pro and V4-Flash. Global Times, citing the company, described a 1 million-token context window, a 384,000-token maximum output, thinking and non-thinking modes, and a Terminal-Bench score of 87.9 against Claude Fable 5’s 88.

Artificial Analysis lists the 0813 checkpoint at 1.6 trillion total parameters with 49 billion active, and peak API prices of $1.32 per million input tokens (cache miss) and $3.96 per million output tokens.

The price change

On August 16 DeepSeek moved from flat rates to peak and off-peak pricing. gHacks, citing eWeek, said some API prices rose between 50% and 1,100% depending on the model and the hour, with off-peak at half of peak. Peak windows cited by Artificial Analysis are 01:00–04:00 and 06:00–10:00 UTC.

• Weights: MIT, Hugging Face, same-day as the changelog
• Serving path in the model card: datacenter-class (example: 4×GB300)
• V4-Flash remains the cheaper sibling (284B total / 13B active in public write-ups)

Takeaways

• GA is real; the April drop was the preview
• Open weights and a price hike landed in the same week
• DeepSeek is positioning V4 Pro as an agent/coding flagship, not a chat toy

Source: gHacks / DeepSeek changelog