TLDR
- V4-Pro model available at 75% reduced pricing through May 5, 2026
- API cache hit pricing reduced by 90% across DeepSeek’s complete model range
- Two V4-Pro configurations available: full Pro edition and streamlined Flash edition
- Model optimized for Huawei semiconductor architecture, demonstrating superior performance against competing open-source solutions
- Strategic pricing reflects escalating competitive dynamics between Chinese and Western artificial intelligence firms
DeepSeek, a Chinese artificial intelligence company based in Hangzhou, has announced dramatic price reductions for its latest V4-Pro model, marking a significant escalation in the global competition among AI developers.
The promotional pricing structure became available to developers in the previous week, with the special rates remaining valid until 15:59 UTC on May 5, 2026.
According to the new pricing framework, cache miss input costs decreased from $1.74 to $0.435. Cache hit inputs now cost $0.03625, down from $0.145, while output charges fell from $3.48 to $0.87.
DeepSeek has additionally implemented a 90% reduction on cache hit input costs throughout its complete API portfolio. This adjustment became effective immediately and is designed to provide substantial savings for developers submitting recurring or similar queries.
The V4-Pro release represents the culmination of extensive development efforts. Notably, the model has been engineered to function with Huawei’s chip infrastructure—a significant consideration as U.S. export controls continue to restrict Chinese enterprises’ access to American-manufactured semiconductors.
Two Versions, One Goal
DeepSeek offers the V4 model in dual configurations. The Pro configuration delivers enhanced capabilities and commanded premium pricing before the discount implementation. The Flash configuration provides a more compact, economical alternative.
According to DeepSeek’s performance metrics, the Pro configuration surpasses competing open-source models in global knowledge evaluation benchmarks. Only Google’s proprietary Gemini-Pro-3.1 achieves higher scores in these assessments.
The company positions the V4 models as purpose-built for AI agent applications. These sophisticated systems execute more advanced operations than conventional chatbot interfaces, though they demand greater computational resources.
This pricing strategy emerges following DeepSeek’s earlier R1 model release, which catalyzed widespread price competition throughout the AI sector upon its debut last year.
A Broader Price War
As AI enterprises transition from experimental phases to production-scale deployment of large language models, reducing inference and operational expenses has emerged as a critical competitive differentiator.
DeepSeek’s aggressive pricing approach is anticipated to compel competitors to implement corresponding reductions, particularly within China’s market, where firms are developing alternatives to Western technologies.
American technology export restrictions have accelerated this transformation, catalyzing the expansion of domestic AI infrastructure across China.
OpenAI, Anthropic, and Google continue releasing advanced models at a rapid pace. Premium pricing for access to these platforms creates opportunities for DeepSeek’s cost-competitive positioning.
The 75% promotional discount on V4-Pro continues through May 5. The comprehensive API pricing adjustments across DeepSeek’s entire model suite are currently operational.
