The company announced that developers integrating its newly unveiled DeepSeek V4-Pro model will benefit from a 75% discount until May 5, a limited-time incentive aimed at boosting early usage and testing across applications. In parallel, DeepSeek is also reducing pricing for input cache hits across its entire API suite to one-tenth of the previous cost, significantly lowering operational expenses for developers relying on repeated query processing.
The announcement was made via a post on X (formerly Twitter), alongside the formal introduction of the V4 model preview, which has been optimized for deployment on hardware developed by Huawei. This alignment with domestic chip architecture reflects ongoing efforts in China’s AI industry to reduce dependence on foreign semiconductor ecosystems.
The new model is being released in two configurations: a higher-performance, premium-tier “Pro” version and a lighter, more cost-efficient “Flash” variant. According to DeepSeek, the Pro model delivers strong results on world-knowledge benchmarks, ranking just behind the closed-source Gemini Pro 3.1 developed by Google, while outperforming several other open-source competitors.
DeepSeek positioned the V4 family as particularly suited for “AI agent” applications—systems designed not just to respond to prompts but to autonomously execute multi-step tasks, interact with tools, and manage more complex workflows. These capabilities typically require higher computational resources, making pricing strategies and efficiency improvements a key competitive lever.
By combining steep pricing reductions with performance-focused model upgrades, DeepSeek appears to be targeting rapid developer uptake in a market increasingly defined by the trade-off between cost, scalability, and model capability. The move also underscores the growing emphasis on agentic AI systems as the next frontier beyond traditional chatbot-style interactions.
As competition intensifies among global AI providers, the V4 launch and accompanying pricing strategy reinforce the broader shift toward aggressively commoditizing model access while differentiating through performance tiers and specialized deployment capabilities.