DeepSeek introduces new API pricing system based on peak and off-peak hours, price increases reach 1100%

After weeks of anticipation regarding price hikes, DeepSeek announced new tariffs for its API on August 13th, coinciding with the official launch of the V4 model family. To allocate resources more rationally, the company is introducing a pricing system divided into peak and off-peak periods, where the cost during off-peak hours is set at half the peak rate. This encourages users to schedule tasks according to actual usage. The new rates take effect at midnight Beijing time on August 17, 2026. DeepSeek had previously presented a peak and off-peak pricing model at the end of June, which many perceived as a hidden price increase.

The price adjustments for V4 are comprehensive. Under the previous pricing structure, V4 Pro input data during off-peak hours with cache hit cost 0.025 yuan per million tokens, input data without cache hit cost 3 yuan per million tokens, and output data cost 6 yuan per million tokens. Under the new tariffs, V4 Pro input data during off-peak hours with cache hit will cost 0.15 yuan per million tokens, without cache hit—4.5 yuan, and output data—13.5 yuan. Analyzing only the changes for off-peak V4 Pro, the smallest increase was 50%, and the largest was 500%.

On August 6th, DeepSeek announced plans for a substantial increase in overall API service prices, and the new tariffs appeared alongside the official release of V4 Pro. Early on August 13th, DeepSeek V4 Pro was officially launched and updated in the API under the version DeepSeek-V4-Pro-0813, which features enhanced reasoning capabilities. According to comparative model tests published by DeepSeek, version V4-Pro-0813 demonstrated improvements in all aspects compared to the previous V4-Pro-Preview. On the DeepSWE benchmark for AI programming agents, the score rose from 7.3 for Preview to 62.7, surpassing even Claude Opus 4.8. Furthermore, on the Cybergym AI security benchmarks and AutomationBench for agents, it outperformed Claude Fable 5, which is considered one of the strongest current models.

This change in pricing policy has industry implications. DeepSeek's V4 model family served as a benchmark for the pricing of low-cost Chinese models, and the company's decision to raise API prices signals a shift in the discussion about internal pricing strategy: the era of sustainable token dumping at low cost may be coming to an end as advanced models approach capabilities, and competition shifts towards application economics.

The new version also strengthens DeepSeek's position as a foundational model underpinning a growing suite of agentic products, as V4-Pro-0813 now provides coding and agent benchmark growth previously held by flagship US models. The timing of this event is also significant: the release of V4 Pro and the price change coincided with the open-sourcing of DeepSeek Harness on the same day, as well as the announcement that Kingsoft's WPS Lingxi Pro would be one of the first products to integrate DeepSeek-V4-Pro. DeepSeek is repositioning itself from a pure model seller to a company monetizing full task completion, where token pricing is just one of several levers of control.

DeepSeek has made its V4 Pro model publicly available, releasing it under the name DeepSeek-V4-Pro-0813 on its pricing page. The flagship model is equipped with a context window of 1 million tokens and a maximum output of 384 thousand tokens, defaulting to a thinking mode, but also providing an endpoint without thinking mode for latency-sensitive requests.

The combination of a 1 million token input and 384 thousand token output allows for processing and generating a significantly larger volume of material within a single request compared to the previous generation, which is particularly relevant for working with large codebases, extensive document corpora, and multi-stage agents.

DeepSeek has also integrated features that most agent developers expect: structured JSON output, tool calling support, Responses API, and compatibility with Anthropic-API. Furthermore, beta support for conversation prefix continuation and FIM completion has been implemented, with FIM limited to the non-thinking mode.

At the service level, endpoints compatible with both OpenAI and Anthropic are offered, allowing teams already using either standard to switch without needing to rebuild their infrastructure.

The pricing reflects a deliberate separation of tiers. The Pro model costs 0.025 yuan per million tokens when cached, 3 yuan for cache misses on input, and 6 yuan on output. In contrast, the Flash model is priced at 0.02, 1, and 2 yuan respectively, making Pro approximately three times more expensive than Flash for input and output cache misses.

Differences are also highlighted by parallelism limitations: Flash supports 2500 concurrent requests, whereas Pro is limited to 500. This ratio indicates a clearer product structure for the V4 family. Flash is geared towards high-frequency, high-throughput traffic where cost dominates, while Pro concentrates power on heavier reasoning tasks, long-context coding, and agent workflows, where the model has a greater performance reserve for use.

Nevertheless, the absolute price of Pro remains moderate by the standards of advanced models, allowing DeepSeek to offset costs after the recent price increase announcement while remaining cheaper than most foreign competitors. Thus, the V4 lineup presents a two-pronged offering: a low-cost, high-throughput Flash tier for scaling, and a more powerful Pro tier aimed at coding and agent tasks.