DeepSeek slashes V4-Pro API prices to 3/6 RMB per million tokens permanently
1.6T parameter model now costs $0.87 per million tokens – permanent price after May 31.
DeepSeek announced a permanent price reduction for its flagship DeepSeek-V4-Pro API model, effective after May 31, 2026. The new rates are 3 RMB per million input tokens and 6 RMB per million output tokens — matching the current 75% promotional discount that expires next week. This means developers will pay roughly $0.87 per million tokens for a 1.6 trillion parameter model, making it one of the cheapest frontier LLM APIs available. The company expects compute expansion and additional cost reductions in the second half of 2026, driven by newer hardware (likely NVIDIA H100/H200 or domestic alternatives).
Industry observers note that DeepSeek is the only lab aggressively pushing inference costs toward zero, even as competitors like OpenAI and Anthropic maintain higher pricing. The move effectively eliminates the "original price" that was only in effect for a few days, prompting social media commentary that "only at DeepSeek" would a discount become the permanent price. For developers and enterprises, this opens the door to large-scale AI applications that were previously uneconomical, especially in China's domestic market where RMB pricing simplifies adoption. The strategy aligns with DeepSeek's stated mission to make intelligence "too cheap to meter," potentially reshaping the competitive landscape in 2026.
- DeepSeek-V4-Pro API pricing permanently reduced to 3 RMB input / 6 RMB output per million tokens after May 31, 2026.
- Current 75% discount ends in ~1 week; new permanent price matches the promotional rate.
- Model has 1.6 trillion parameters, offering inference at ~$0.87 per million tokens, among cheapest for frontier models.
Why It Matters
DeepSeek's aggressive pricing pressures competitors and makes cutting-edge AI inference accessible at near-zero cost.