DeepSeek slashes V4 Pro price 75%

DeepSeek has made the 75% discount on its V4-Pro model permanent, locking in a price cut that had been scheduled to expire at the end of May. The company’s pricing page now shows V4-Pro at $0.003625 per million cached input tokens, $0.435 per million input tokens and $0.87 per million output tokens. DeepSeek’s documentation says the prices listed are per 1 million tokens and that the V4-Pro rate will be “officially adjusted to 1/4 of the original price” after the promotion ends on May 31, 2026. ### How big is the cut in actual posted prices? DeepSeek’s own pricing table shows the old V4-Pro rates alongside the discounted ones. Cached input is listed at $0.003625 per million tokens versus $0.0145 before, input at $0.435 versus $1.74, and output at $0.87 versus $3.48. The same page says the cache-hit price for all models was reduced to one-tenth of launch pricing effective April 26, 2026. (api-docs.deepseek.com) Reuters reported on May 23 that DeepSeek cut V4-Pro API costs to between 0.025 yuan and 6 yuan per million tokens, down from 0.1 yuan to 24 yuan previously, depending on usage type. Reuters said the company described the move in a statement issued Saturday. ### Why does cached input matter so much here? (api-docs.deepseek.com) DeepSeek’s pricing page separates “cache hit” from “cache miss” input, which means repeated prompt material can be billed at a much lower rate than fresh input. At $0.003625 per million tokens for V4-Pro cache hits, the cost of reusing long system prompts, documents or repeated context falls far below the standard input rate of $0.435 per million. (money.usnews.com) The practical effect is clearest in applications that resend large blocks of unchanged context. Teams running retrieval pipelines, coding copilots or agent workflows often reuse instructions, tool schemas and prior conversation state; under DeepSeek’s posted rates, that repeated context is now billed at a fraction of even the already reduced base input price. That follows directly from the company’s published cache-hit and cache-miss schedule. (api-docs.deepseek.com) ### What did DeepSeek say about why V4-Pro had been expensive before? When DeepSeek launched V4 last month, Reuters said the company had told users the Pro version would cost up to 12 times more than the less powerful Flash version because of “constraints in high-end compute capacity.” Reuters also reported that DeepSeek had said Pro pricing was expected to fall once Huawei Ascend 950 supernodes were launched in large quantities in the second half of the year. (api-docs.deepseek.com) Reuters said DeepSeek did not disclose whether the permanent cut was tied to increased supply of Huawei’s Ascend 950 chips, which the company used to maximize V4 performance. Reuters added that U.S. export controls have helped Huawei’s AI chip sales in China while other controls have limited Huawei’s ability to scale Ascend production. (money.usnews.com) ### What does the pricing page say happens next? DeepSeek’s documentation says the 75% V4-Pro discount promotion ends on May 31, 2026 at 15:59 UTC, after which the model’s API pricing will be officially adjusted to one-quarter of the original price. The page also says product prices may vary, that DeepSeek reserves the right to adjust them, and that users should “regularly check” the pricing page for the most recent rates. (money.usnews.com) Reuters said DeepSeek launched V4 in April 2026. The next concrete milestone in DeepSeek’s own materials is the May 31, 2026 end of the promotional period, with the same reduced V4-Pro rates remaining on the published schedule after that date. (api-docs.deepseek.com)