Details
- DeepSeek announced DeepSeek-V4-Pro, adding major agent improvements aimed at stronger production performance.
- The V4 lineup now includes flexible reasoning effort settings for V4-Pro and V4-Flash, ranging from low for simple tasks to max for complex agent workflows.
- DeepSeek said V4-Pro includes native OpenAI Responses API support and is optimized for that interface.
- The company is also changing API billing to a peak and off-peak model, with off-peak rates set at 50% below peak.
- New pricing takes effect at 16:00 UTC on August 16, 2026, giving users a short window before the updated rate card applies.
- DeepSeek’s own pricing docs for the current V4 lineup show a lower baseline schedule today, so this announcement represents a clear pricing and product reset for the V4 family.
- The timing and rate structure suggest DeepSeek is encouraging customers to schedule heavier workloads outside peak usage windows.
Impact
DeepSeek is pairing a model upgrade with a more granular pricing scheme, which could make the V4 line easier to adopt for teams that want to control inference costs and route workloads by urgency. The native Responses API support also makes the model more compatible with OpenAI-style agent stacks, reducing integration friction for developers who want a lower-cost alternative without redesigning workflows. The new peak and off-peak structure may improve capacity management for DeepSeek while nudging customers toward batchable, cost-sensitive usage.
