DeepSeek cuts V4 Pro price 75%

DeepSeek said on May 23 that it will permanently keep a 75% discount on its flagship V4-Pro model, locking in pricing that had previously been billed as a temporary promotion. The Chinese AI startup’s pricing page says V4-Pro output tokens will cost $0.87 per million and input tokens $0.435 per million after the offer’s scheduled end on May 31. The move was first reported by Bloomberg and The Next Web, which said the lower rates intensify price competition across the model market. DeepSeek’s documentation says the V4-Pro price will be adjusted to one-quarter of its original level after the promotion ends. ### Which price did DeepSeek actually lock in? DeepSeek’s API documentation says V4-Pro output tokens are priced at $0.87 per million and input tokens at $0.435 per million under the revised schedule. The same page says cached input pricing was also reduced earlier, to one-tenth of the launch price for all models. The company’s documentation lists the previous V4-Pro range at up to $3.48 per million output tokens and $1.74 per million input tokens, implying a 75% cut. Bloomberg reported on May 23 that DeepSeek would make the discount permanent rather than let it expire at the end of May. ### When does the permanent pricing take effect? DeepSeek’s pricing page says the 75% discount promotion ends on May 31, 2026, at 15:59 UTC, and that V4-Pro pricing will then be “officially adjusted” to one-quarter of the original price. That means the promotional rate and the post-promotion rate are the same in practice, based on the company’s published schedule. The Next Web reported on May 24 that DeepSeek had made the discount permanent, citing the company’s website. Engadget reported the same day that the update converted what had been a temporary promotion into a standing price change. ### How far below rival pricing does that put V4-Pro? The Next Web said DeepSeek’s new V4-Pro rate sits below pricing it cited for GPT-5 and Gemini in its comparison. The article described the move as an escalation in the AI price war as model vendors compete for developers and enterprise workloads. Bloomberg said the cut would intensify competition between Chinese AI firms and global peers. DeepSeek has been one of the most aggressive vendors on price since entering the market with lower-cost models aimed at developers. ### Why does this matter for developers using agents? Long-running agent workloads can generate large output volumes, making output-token pricing a direct operating cost. The Next Web said the new V4-Pro rate could affect model-routing decisions for cost-sensitive tasks because output-heavy workflows, retries and tool loops can raise token bills quickly. DeepSeek’s own pricing page also highlights lower cache-hit costs across its API lineup. For developers running repeated prompts or retrieval-heavy systems, those lower cached-input rates can matter alongside the headline output price. ### What should readers watch next? May 31 at 15:59 UTC is the next concrete milestone on DeepSeek’s pricing page. Rival responses, if any, are likely to show up first on official pricing pages from OpenAI, Google and Anthropic, which developers use to compare token costs and routing choices.