DeepSeek Permanently Slashes V4-Pro Model Prices by 75%

In a bold move that is set to significantly escalate the ongoing global artificial intelligence price war, Chinese AI powerhouse DeepSeek has declared a permanent 75% price reduction for its cutting-edge V4-Pro model. This decision transforms what was initially a temporary promotional discount, slated to conclude at the end of May, into a standing pricing policy, cementing DeepSeek's position as a highly competitive and cost-effective provider in the generative AI landscape. The new, drastically reduced rates position DeepSeek V4-Pro as an exceptionally attractive option for developers and enterprises, with typical API pricing now hovering around US $0.435 per million input tokens. This makes the V4-Pro model more than 20 times cheaper than equivalent offerings from major Western AI developers such as Google, OpenAI, and Anthropic. The implications of this permanent price cut are far-reaching, particularly for Chief Information Officers (CIOs) and organizations grappling with escalating token expenditures for their AI workloads. For high-volume users engaged in compute-intensive tasks like Retrieval Augmented Generation (RAG) pipelines, sophisticated code review agents, or extensive long-context inference, switching to DeepSeek V4-Pro could translate into substantial annual savings, potentially reaching millions of dollars. This shift in the cost calculus for enterprise AI workloads is expected to prompt many teams to re-evaluate their current model choices, especially if they are defaulting to more expensive closed-source alternatives. According to industry analysts, including Sanchit Vir Gogia of Greyhound Research, DeepSeek's ability to implement and sustain such aggressive price reductions is not merely a marketing ploy or a strategy to capture market share, although it certainly serves those purposes. Instead, it is fundamentally underpinned by significant engineering advancements in the V4-Pro model itself. The model has been meticulously engineered to drastically reduce its compute and memory footprint, allowing for a much more efficient operation at a lower cost base. This efficiency gain enables DeepSeek to offer frontier-class coding and reasoning capabilities at a fraction of the price of its rivals. DeepSeek V4-Pro boasts impressive performance metrics, scoring 80.6% on SWE-bench Verified and operating with a Mixture-of-Experts (MoE) architecture comprising 1.6 trillion total parameters, with 49 billion active parameters per forward pass. Crucially, the model is released under an MIT license, granting full commercial use rights, and supports a 1-million-token context window. This combination of high performance, open weights for self-hosting, and unprecedented affordability presents a compelling value proposition. For organizations prioritizing data sovereignty or seeking to run AI systems on-premise, the availability of open weights under a permissive license, coupled with the low API costs, offers a powerful alternative to proprietary models. The permanent nature of these price cuts signals a long-term commitment from DeepSeek to disrupt the AI market, placing considerable pressure on competitors to innovate on both performance and pricing.