# DeepSeek Raises API Prices Up to 11-Fold Just Months Before a Rumored IPO > Source: > Published: 2026-08-17 03:31:14+00:00 *DeepSeek just raised its own API prices by as much as 11 times, ending the era of dirt-cheap Chinese AI that reset the entire industry's pricing eighteen months ago.* The lab that broke OpenAI's pricing model in January 2025 has now broken its own. DeepSeek's V4 price increase took effect at 16:00 UTC on Sunday, August 16. It isn't subtle. Output tokens for V4-Flash, the cheap workhorse model developers use for high-volume jobs, jumped from $0.28 to as much as $1.32 per million tokens during peak hours, according to figures confirmed by Bloomberg and InfoWorld. V4-Pro's output price nearly quadrupled, from $0.87 to $3.96 per million tokens at peak. Input pricing moved too: V4-Flash input rose from $0.14 to $0.44 per million tokens at peak, and V4-Pro input roughly tripled to $1.32. That's not a rounding adjustment. That's a company changing its mind about who pays for its compute. DeepSeek warned developers the hike was coming on August 6, giving them ten days to prepare. Its explanation, relayed by Qz and InfoWorld, is that it's allocating resources more reasonably by pushing usage toward off-peak hours, defined as 01:00 to 04:00 and 06:00 to 10:00 UTC. Peak-hour rates now run double the off-peak ones, and in some cache-hit categories the increase reaches 1,100%, InfoWorld's breakdown of the new rate card shows. If you built a product around DeepSeek's rock-bottom rates, you now have a choice. Reschedule your batch jobs for the middle of the night in Beijing, or eat the cost. [The US government just gave DeepSeek a quiet pass while blacklisting more than 100 other Chinese firms](https://startupfortune.com/the-us-government-just-gave-deepseek-a-quiet-pass-while-blacklisting-more-than-100-other-chinese-firms/) The Trump administration suspended plans to add DeepSeek to its commercial blacklist, even as it designated more than 100 other Chinese firms as national security risks. The decision coincides with DeepSeek closing a $7.4 billion funding round that handed China's national AI fund direct voting rights over the lab, signaling a deliberate leverage... ## How DeepSeek broke the market in the first place DeepSeek's original disruption is well documented at this point. When it released R1 on January 20, 2025, the model cost roughly $0.55 per million input tokens, about 27 times cheaper than OpenAI's o1 at the time. That single price point rattled Silicon Valley badly enough that OpenAI cut its own API prices and leaned harder into smaller mini models, Google answered with Gemini 2.0 Flash, and Anthropic later cut Claude pricing by roughly 67% in one announcement. DeepSeek didn't just undercut the market. It forced a repricing of the entire industry. Eighteen months on, DeepSeek is learning what every company that wins on price eventually learns: cheap is expensive to run. Serving V4 at scale on compute constrained by U.S. export controls, while demand keeps climbing, costs real money. Something had to give, and DeepSeek chose to give up its cheapest offer rather than its uptime. There's also the matter of what comes next for the company. DeepSeek is reportedly mid-fundraise and weighing an IPO, a detail Bloomberg and other outlets have tied to founder Liang Wenfeng's broader strategy for the business. Nobody at DeepSeek has said the price hike is about satisfying future investors rather than server bills. But the timing is hard to ignore. ## What happens now For a year, "too cheap to be true" was the working assumption about Chinese AI pricing, and DeepSeek just confirmed it. Rivals who refused to race DeepSeek to zero, OpenAI with its tiered mini models, Anthropic with Claude's enterprise focus, look smarter this week than they did last month. That doesn't mean the price war is over. It means DeepSeek just admitted it was losing money on the version everyone loved. Fortune and Semafor both noted that the increase lands as Western labs face their own pressure to justify AI infrastructure spending, which makes DeepSeek's retreat from its own headline price feel less like an isolated move and more like an industry-wide reckoning with what inference actually costs. For developers, the fix is boring but real. Shift batch inference and anything that isn't latency-sensitive into DeepSeek's off-peak window, and start re-running the numbers against OpenAI's and Anthropic's smaller models, because the gap that made DeepSeek the easy default just got a lot narrower. [DeepSeek is making its 75 percent API discount permanent](https://startupfortune.com/deepseek-is-making-its-75-percent-api-discount-permanent/) DeepSeek has made its 75 percent V4-Pro API discount permanent after the promotion period, resetting the cost expectations for AI startups and developers. The move puts pricing pressure on OpenAI, Anthropic and Google while raising fresh questions about DeepSeek’s margin strategy as it pursues a major funding round. **Also read:** [Andrew Ng names the four AI skills that decide which teams actually ship](https://startupfortune.com/andrew-ng-names-the-four-ai-skills-that-decide-which-teams-actually-ship/) • [Stripe Agrees to Buy AI Router OpenRouter for More Than $7 Billion](https://startupfortune.com/stripe-agrees-to-buy-ai-router-openrouter-for-more-than-7-billion/) • [Nvidia's $40 Billion Spending Spree Revives Dot-Com Bubble Fears](https://startupfortune.com/nvidias-40-billion-spending-spree-revives-dot-com-bubble-fears/)