Chinese AI startup DeepSeek is making its 75% price cut on the flagship DeepSeek-V4-Pro model permanent, keeping API prices at one-fourth of the original rate as competition intensifies across open and closed AI models. The move could put fresh pressure on global AI providers competing on cost, performance, and developer adoption.
TL;DR
- DeepSeek will permanently keep V4-Pro prices at 25% of the original level after its promotion ends.
- The official pricing page lists V4-Pro at $0.003625 per 1 million cache-hit input tokens, $0.435 for cache-miss input, and $0.87 for output.
- The model supports a 1 million-token context window and up to 384,000 output tokens.
- Reuters linked the cut to questions around Huawei Ascend chip availability, though DeepSeek did not confirm the reason.
DeepSeek is making its flagship AI model a lot cheaper, permanently. According to Reuters, the China-based artificial intelligence startup will keep a 75% price cut on its DeepSeek-V4-Pro model after the discount period ends. The move means the model’s API pricing will remain at a quarter of its original level, rather than snapping back to launch pricing.
DeepSeek’s own API pricing page confirms the change. It says the DeepSeek-V4-Pro model API pricing will be officially adjusted to 1/4 of the original price after the 75% promotion ends on May 31, 2026, at 15:59 UTC.
The updated listed prices show V4-Pro at $0.003625 per 1 million input tokens for cache hits, $0.435 per 1 million input tokens for cache misses, and $0.87 per 1 million output tokens.
That is a meaningful shift for developers and enterprises building AI agents, coding tools, long-context workflows, and other token-heavy applications.
DeepSeek-V4-Pro is not positioned as a lightweight chatbot model. The company describes it as a 1.6 trillion total parameter model with 49 billion active parameters, while its cheaper V4-Flash model has 284 billion total parameters and 13 billion active parameters.
DeepSeek says both models support a 1 million-token context window, with maximum output of up to 384,000 tokens.
The company has also pushed V4-Pro as a strong model for agentic coding, reasoning, math, STEM, and world knowledge tasks. In its V4 preview release, DeepSeek claimed the Pro model rivals top closed-source models and leads current open models in several areas, while V4-Flash is positioned as the faster and more economical option.
The pricing decision also comes with a bigger geopolitical and infrastructure backdrop.
Reuters reported that DeepSeek did not disclose whether the permanent cut was due to increased supply of Huawei’s Ascend 950 chips, which the company has used to maximize V4’s performance.
Reuters also noted that Huawei’s AI chip sales have benefited from U.S. export controls that limit Nvidia’s ability to sell its most advanced semiconductors in China, though restrictions on chipmaking equipment have also limited Huawei’s ability to scale Ascend production.
When V4 was launched in April, DeepSeek said the Pro version would cost up to 12 times more than the Flash version because of constraints in high-end compute capacity. It also said Pro pricing was expected to fall sharply once Huawei Ascend 950 supernodes were launched in larger quantities in the second half of the year.
This update makes that expected drop official, and earlier than many price-sensitive AI users may have expected.
Reuters had previously reported that DeepSeek offered developers a 75% discount on V4-Pro after launching the V4 preview in April. At the same time, the company cut input cache-hit prices across its API lineup to one-tenth of the original price.
Topics for more insights:
For the wider AI market, the signal is clear. Model providers are no longer competing only on benchmark scores, context windows, or coding ability. They are also competing on whether developers can afford to run complex AI workloads at scale.
South China Morning Post reported that V4-Pro topped a global bang-for-buck ranking after the 75% cut, while Engadget noted that the lower prices could be aimed at undercutting competition.
The cut may also influence how businesses evaluate open-weight and China-based AI models against U.S. rivals. If DeepSeek can combine strong reasoning performance, long-context support, agentic capabilities, and aggressive pricing, it could make the economics of AI deployment tougher for competitors that charge more for comparable workloads.
However, there are still questions. DeepSeek’s pricing page also says product prices may vary and recommends users regularly check the page for the latest pricing. That means the current reduction is permanent relative to the promotion ending, but customers should still monitor future updates.
For now, DeepSeek has made its move. The company is taking a flagship model that was once constrained by high-end compute capacity and making it dramatically cheaper to run.
In an AI market where every token counts, that could be enough to force the next round of price cuts.
