Model launches

AI-generated text

DeepSeek launches V4 Pro with higher performance and premium pricing

Chinese AI company DeepSeek introduced its V4 Pro model, positioning it as a premium offering with substantially higher per-token prices than the V4 Flash.

DeepSeek launches V4 Pro with higher performance and premium pricing

Chinese AI firm DeepSeek on Thursday officially introduced its V4 Pro language model, positioning the new model as a premium offering compared with the V4 Flash. The company said the V4 Pro will be available via API, mobile app and web interface.

Pricing and billing changes

According to data from independent analyst firm Artificial Analysis, the DeepSeek V4-Pro-0813 model is priced at $1.32 per million input tokens and $3.96 per million output tokens. Those rates are roughly nine times and fourteen times higher, respectively, than the V4 Flash’s unit prices. DeepSeek also announced an increase in API prices for both V4 Pro and V4 Flash and introduced peak and off-peak pricing.

Performance and independent testing

Notably, last month’s V4 Flash outperformed a preview release of V4 Pro in several independent tests, suggesting rapid development between the preview and final releases. The officially published V4 Pro, however, represents a measurable improvement in independent evaluations: Artificial Analysis assigned the V4 Pro reasoning variant a score of 53 on its intelligence index versus 40 for V4 Flash. The index assesses nine capabilities, including agent tasks, tool use, coding, scientific reasoning and long-context handling.

Market context and strategic plans

DeepSeek became a major player in China’s AI sector in early 2025 with wide adoption of its R1 model, challenging the notion that high-performance AI systems must be built at the cost level of U.S. competitors. Since then, Chinese rivals such as Moonshot AI, Zhipu AI, MiniMax, Alibaba and ByteDance have intensified competition with successive model launches.

The company is focusing on turning market recognition into a sustainable business model: Reuters reported in July that DeepSeek is planning further fundraising at an implied valuation of about $74 billion, after raising roughly $7.4 billion in its first external funding round in June. Plans include at least doubling headcount — particularly in data-center and AI-agent development teams — and developing an in-house AI chip to reduce dependence on Nvidia and Huawei.

Why this matters

The V4 Pro launch and the associated pricing moves indicate DeepSeek’s push toward higher revenue and sustainability, supported by independent performance gains. At the same time, higher prices and new billing structures will be significant considerations for users and enterprise customers, especially when scaling API usage.

(Source: Reuters; data: Artificial Analysis)