DeepSeek flipped V4-Pro to general availability on Aug. 13 and pitched it as an agent-first upgrade, adding native OpenAI Responses API support with one-click Codex setup and a tri-tier reasoning-effort switch (low, high, max). The same rollout confirmed a new peak/off-peak API price schedule that lifts rates by as much as 1,100% versus the previous V4 tariff, taking effect at 16:00 UTC on Aug. 16.
What V4-Pro Actually Ships
V4-Pro is now live on DeepSeek's app and web via “Expert Mode,” and is available through the API under the same model name as before. The company is pointing customers at flexible reasoning effort — low for simple work, high for daily agent flows, max for complex tasks — and is highlighting production-grade gains for agentic pipelines rather than a benchmark headline number.
End Of The Rock-Bottom Era
The new pricing schedule prices V4-Pro at RMB 0.025 per million cached input tokens, RMB 3 per million uncached input tokens and RMB 6 per million output tokens at peak, with off-peak windows offering a 50% discount to encourage workload scheduling. Caixin flagged the change as a jump of up to 1,100% versus V4-Flash pricing depending on the token category and time of use — a clear signal that DeepSeek believes premium reasoning workloads are inelastic enough to bear the mark-up.
Reshuffling China's Model Economics
V4-Pro lands as U.S. rivals cut prices to defend enterprise share: compare with our coverage of Google's Gemini 3.7 Flash at half price, OpenAI's Ultrafast tier on Cerebras, and Z.ai's GLM-5.3 open-weights release. Cheap Chinese inference was DeepSeek's opening act; V4-Pro is the pivot from disruption to monetisation.
Reporting based on coverage from DeepSeek API Docs, Caixin, Reuters and TechStartups.