DeepSeek v4 full release set for mid-July with peak-hour pricing doubled
DeepSeek notified users that v4 will exit preview in mid-July with doubled peak-hour API rates, while off-peak pricing holds steady.
DeepSeek notified users this week that v4, currently available as a preview, will receive its full release in mid-July. The email outlined a pricing change that doubles API costs during peak hours, while rates outside two daily windows—one lasting three hours, the other four—remain unchanged. The move marks a shift from the flat-rate preview period and signals DeepSeek's intent to manage capacity as demand scales.
The current v4 preview has drawn attention for its reasoning capabilities and competitive per-token rates, positioning it alongside other open-weight and API-driven models in the sub-$1-per-million-token tier. Users running production workloads on the preview have been advised to budget for the new rate structure.
Peak pricing and off-peak windows
Under the new model, peak-hour API calls will cost twice the current rate. DeepSeek defined two off-peak windows totaling seven hours per day where pricing stays flat. The company did not specify the exact UTC boundaries of those windows in the user email, leaving developers to wait for official documentation. The tiered approach mirrors strategies used by cloud providers to smooth load, though the doubling during peak periods is steeper than typical compute surcharges.
Some users hope the full release will add native multimodal input—image and audio alongside text—a feature the preview does not support. DeepSeek has not confirmed whether v4 will ship with multimodal capabilities at launch, but the absence of that feature in the preview has been a frequent point of feedback. Competitors including Qwen and several open-weight alternatives already offer vision and audio modalities, making multimodal support a logical next step for DeepSeek's roadmap.
