DeepSeek cuts V4-Pro prices by 75% and makes cache hits ten times cheaper across its API

DeepSeek has sharply escalated its price attack on the AI API market: the company announced a temporary 75% discount on DeepSeek-V4-Pro and, at the same time, cut the cost of cache hits tenfold across its entire API lineup. For developers, this is not just a promotion running until May 5, 2026, but an attempt to make switching to a Chinese model nearly painless financially, even for teams already working with OpenAI, Anthropic or Google.

According to DeepSeek's current price table, the standard V4-Pro rate is $1.74 per million input tokens on a cache miss, $0.0145 per million cached input tokens and $3.48 per million output tokens. The temporary promotion, valid until 15:59 UTC on May 5, 2026, lowers these figures to $0.435, $0.003625 and $0.87 respectively. At the same time, the company updated its caching rules for the whole API lineup: the cache hit price is now one tenth of its original launch level. This matters especially for production agents, because they constantly reuse the same system instructions, long prefixes and chunks of context.

The move looks strategic, not just a marketing play. DeepSeek has pressured the market on price before, especially after the R1 release in January 2025, when it became clear that the Chinese company was ready to compete not only on quality but also on the cost of running its models. Now the stakes are higher: V4-Pro launched on April 24, 2026, and on April 27 the company announced the aggressive API discount.

Against this backdrop, the offer looks like a direct challenge to American providers, which have gradually cut prices themselves in recent months, though not nearly as sharply. The political backdrop adds to the effect: the administration of Donald Trump is simultaneously accusing Chinese AI companies of large-scale distillation of American models.

V4-Pro is not built for price competition alone. According to DeepSeek, it is a mixture-of-experts model with 1.6 trillion total parameters and 49 billion active parameters per task. It supports a context window of 1 million tokens and a maximum output of up to 384 thousand tokens, which makes it considerably more practical for long documents, large codebases and multistep agentic workflows. The company also stresses that the API is compatible with familiar OpenAI and Anthropic formats, and that it integrates natively with Claude Code, OpenClaw and OpenCode. That lowers the cost not only of usage but of migration itself: there is no need to change your whole stack for a new model.

Another important layer of the story is infrastructure. DeepSeek promotes V4 as a model optimized for Chinese Huawei Ascend 950 chips and Cambricon hardware, not only for Nvidia. For the market, this signals that competition is no longer just between individual models but between entire technology stacks: their own accelerators, their own API layer, their own agentic tools and their own pricing policy. If this combination really delivers stable quality at long context, the pressure on closed American vendors will become systemic rather than episodic.

The takeaway is simple: DeepSeek is trying to win not a single news week but a share of real-world development. When a model has open weights, a million tokens of context, compatibility with popular SDKs and, at the same time, pricing that sharply cuts the cost of repeated requests, there are fewer arguments for staying on a more expensive API. For startups and small teams, this is a chance to launch agentic products more cheaply; for large players, it is a reminder that the next phase of AI competition will be decided not only by answer quality but by the price of each working request.