DeepSeek fixes $0.88 output price

DeepSeek’s published API pricing now shows deepseek-v4-pro at $0.87 per 1 million output tokens after its current promotional period ends on May 31, according to the company’s documentation. The pricing page says the model’s output rate is currently marked “75% off” from $3.48 and that, when the promotion ends at 15:59 UTC on May 31, the API price will be “officially adjusted to 1/4 of the original price.” A May 24 social-media post amplified the change as a permanent price fix for V4 Pro and described it as disruptive to prevailing model pricing. DeepSeek’s own pricing page does not use that language, but it does confirm the underlying numbers and the effective date. ### What, exactly, did DeepSeek lock in? DeepSeek’s pricing table lists deepseek-v4-pro output at $0.87 per 1 million tokens, input cache-miss pricing at $0.435, and input cache-hit pricing at $0.003625, all marked as 75% off the original levels. (api-docs.deepseek.com) A note directly beneath the table says the deepseek-v4-pro API price will be adjusted to one-quarter of the original price after the promotion ends on May 31. The company’s documentation therefore indicates the promotional price becomes the standing listed price after May 31 rather than snapping back to the prior $3.48 per 1 million output tokens. That is the key factual change reflected in the pricing page. ### Why are people focusing on the output-token number? (api-docs.deepseek.com) The $0.87 figure applies to output tokens, the part of the bill that grows when a model produces long answers, code, or agent traces. DeepSeek’s pricing page bills separately for input cache hits, input cache misses and output tokens, and says charges are calculated by multiplying token counts by the listed rates. (api-docs.deepseek.com) The company also says all listed prices are per 1 million tokens and recommends users check the page regularly because product prices may vary and DeepSeek “reserves the right to adjust them.” That means the current documentation confirms the posted rate, while also leaving open the possibility of later revisions. (api-docs.deepseek.com) ### Which model does this apply to? DeepSeek’s current API documentation names deepseek-v4-pro and deepseek-v4-flash as the active model options in its OpenAI- and Anthropic-compatible API formats. The same quick-start page shows a sample call using deepseek-v4-pro with thinking enabled. The pricing page lists deepseek-v4-pro with a 1 million-token context length, a maximum output of 384,000 tokens and a concurrency limit of 500. (api-docs.deepseek.com) Deepseek-v4-flash is listed separately at lower posted rates, including $0.28 per 1 million output tokens. ### Did DeepSeek say anything else about model naming and migration? (api-docs.deepseek.com) DeepSeek’s documentation says the legacy model names deepseek-chat and deepseek-reasoner will be deprecated on July 24, 2026. For compatibility, the company says those names correspond to the non-thinking and thinking modes of deepseek-v4-flash, respectively. (api-docs.deepseek.com) That leaves deepseek-v4-pro as a separately named higher-tier model on the current pricing page, with the post-May 31 output rate shown at $0.87 per 1 million tokens. ### What happens next, and where can users verify it? May 31, 2026 at 15:59 UTC is the next dated milestone on DeepSeek’s pricing page for V4 Pro. (api-docs.deepseek.com) July 24, 2026 is the next dated milestone on the quick-start page for retirement of the deepseek-chat and deepseek-reasoner model names. DeepSeek directs users to its “Models & Pricing” page for current rates and says customers should check that page regularly for the most recent pricing information. (api-docs.deepseek.com)