End of the Free Ride: DeepSeek Releases V4-Pro and Sharply Raises API Prices

The era of Chinese dumping has ended. On August 15-16, 2026, the DeepSeek lab officially launched its flagship agentic model, `DeepSeek-V4-Pro`. The architecture boasts a monstrous 1-million-token context window and is designed for the autonomous execution of complex, multi-step engineering tasks.

However, the main shock to the market lies in the pricing. Starting August 16, the company is raising inference tariffs to $3.96 per million output tokens during peak hours. Having won the battle for audience share with ultra-budget releases, DeepSeek is shifting to strict monetization. The corporate sector, which had migrated en masse to Asian APIs to save on OPEX, is caught in a trap. It is becoming obvious: maintaining Agentic AI infrastructure costs colossal amounts of money, and no lab in the world can subsidize electricity bills for its B2B clients forever.

Source: DeepSeek / Quartz / Global Times