# DeepSeek sets peak V4 API rates at twice off-peak levels as prices rise > Source: > Published: 2026-08-13 11:44:29+00:00 [Liang Wenfeng](https://techcrunch.com/2025/01/28/who-is-liang-wenfeng-deepseek-founder-comes-from-ai-investing/?ref=runtimewire)'s [DeepSeek](https://deepseek.com/?ref=runtimewire) will raise API prices for its V4 models on August 16 while introducing a two-tier rate card designed to move flexible computing jobs outside its busiest hours. DeepSeek said off-peak rates will sit 50% below peak pricing. That framing describes the gap between the two new tiers. Every off-peak rate in the new table remains higher than the corresponding price DeepSeek charges today. [DeepSeek on X](https://x.com/deepseek_ai/status/2087864589895798968?ref=runtimewire) The [new prices](https://x.com/deepseek_ai/status/2087864589895798968/photo/1?ref=runtimewire) take effect at 16:00 UTC on August 16, 2026. DeepSeek defines peak hours as 01:00-04:00 UTC and 06:00-10:00 UTC, totaling seven hours each day. The other 17 hours are off-peak. ### The cheapest window still costs more DeepSeek currently charges V4-Flash customers $0.0028 per 1 million cache-hit input tokens, $0.14 per 1 million cache-miss input tokens and $0.28 per 1 million output tokens. Under the new off-peak schedule, those prices rise to $0.007, $0.22 and $0.66, respectively. That amounts to increases of 150% for cache hits, 57% for cache misses and 136% for output, even when a customer confines its traffic to the cheaper 17-hour window. Peak V4-Flash rates will reach $0.014 for cache-hit input, $0.44 for cache-miss input and $1.32 for output. The increase is steeper for some V4-Pro usage. As of August 13, DeepSeek's [official pricing documentation](https://api-docs.deepseek.com/quick_start/pricing?ref=runtimewire) lists current V4-Pro rates of $0.003625 for cache-hit input, $0.435 for cache-miss input and $0.87 for output per 1 million tokens. The table combines that current schedule with the rates announced for August 16: | Model and period | Cache-hit input | Cache-miss input | Output | | V4-Flash current | $0.0028 | $0.14 | $0.28 | | V4-Flash off-peak | $0.007 | $0.22 | $0.66 | | V4-Flash peak | $0.014 | $0.44 | $1.32 | | V4-Pro current | $0.003625 | $0.435 | $0.87 | | V4-Pro off-peak | $0.022 | $0.66 | $1.98 | | V4-Pro peak | $0.044 | $1.32 | $3.96 | All figures are per 1 million tokens. V4-Pro's off-peak cache-hit price will be roughly six times its current level. Its off-peak cache-miss and output rates rise by about 52% and 128%. During peak hours, a V4-Pro cache hit will cost just over 12 times the current rate. For a workload distributed evenly across all 24 hours, the blended output price would be about three times today's rate for either model. Actual bills will depend on when traffic arrives, whether prompts hit DeepSeek's cache and how much output each request generates. The rate card favors batch evaluation, document processing, data enrichment and other jobs that can sit in a queue. Consumer chat products and autonomous agents have less control over timing. An agent can also turn one user instruction into repeated model calls, magnifying a per-token increase across planning, tool use and retries. RuntimeWire examined that exposure when [DeepSeek cut V4-Pro pricing in July](/article/deepseek-v4-pro-price-cut-agent-margin-problem). ### Liang applies scheduling economics to inference The two-tier structure fits Liang's route into AI. He studied information engineering at Zhejiang University before building algorithms for stock selection and co-founding the quantitative hedge fund High-Flyer, which became DeepSeek's financial and technical base. That background included costly computing infrastructure. High-Flyer's [history](https://www.high-flyer.cn/en/history/?ref=runtimewire) says its Fire-Flyer I cluster cost approximately 200 million yuan and used 1,100 accelerator cards. DeepSeek's new rate card gives developers a reason to move discretionary work into the cheaper window while charging more for traffic that arrives during the seven peak hours. The policy can spread flexible demand across the day without setting every request at the highest rate. Because DeepSeek's peak windows fall outside most normal US working hours, many US-based teams can schedule daytime batch jobs at off-peak rates. Round-the-clock applications will still encounter both tiers. With the new rates taking effect at 16:00 UTC on August 16, developers have three days to change queueing systems, set time-based routing policies or revise customer-facing usage limits. ### V4's low-cost pitch gets a new denominator DeepSeek's documentation lists V4-Flash and V4-Pro and shows the current per-million-token prices used in the comparison. DeepSeek's pricing documentation lists a 1 million-token context window, a maximum output of 384,000 tokens and OpenAI-compatible and Anthropic-compatible API formats. The new rate card changes those comparisons. Benchmark cost figures produced under the current schedule will not describe API economics after August 16. A reproducible cost claim will need to specify the model version, cache behavior and time window alongside token consumption. The pricing table gives developers the rates and effective date. It preserves a cheaper lane for workloads that can be scheduled around demand while charging substantially more for traffic that cannot wait.