DeepSeek and Xiaomi have announced significant price cuts on their AI models, while U.S. companies are raising prices, highlighting a growing disparity.
DeepSeek and Xiaomi Implement Significant Price Cuts
DeepSeek has permanently reduced the price of its V4-Pro model by 75%, now costing just $0.87 per million tokens. This change took effect on May 22 and aims to enhance accessibility to AI services. Similarly, Xiaomi has slashed prices for its MiMo-V2.5 API by up to 99% as of May 26, with the Pro model's cached input priced at $0.0036 per million tokens.
In stark contrast, U.S. firms are increasing costs. OpenAI's newly launched GPT-5.5 now charges $30 per million tokens, while Anthropic's Claude Opus 4.7 has introduced a new tokenizer that may raise costs by 35% despite maintaining current pricing levels.
Understanding Token Pricing
For context, AI models like ChatGPT and Claude typically operate under a subscription model for consumers. However, developers using these APIs face costs based on token usage, where each token roughly equates to three-quarters of a word. This pricing structure is critical for businesses utilizing AI technology.
Xiaomi's recent adjustments provide users with significantly more tokens for the same amount of money, with the Max plan now offering access to 82 billion tokens for $100, a massive increase from the previous 1.6 billion tokens.
Efficiency Innovations Driving Cost Reductions
Fuli Luo, head of Xiaomi's MiMo team, explained that the substantial price cuts are due to advancements in their data storage and processing capabilities. The new system can retain five times more data, which reduces computing power requirements and leads to an 80% drop in storage and processing costs. "Our deepest price reduction, reaching up to 99%, applies specifically to Input (Cache Hit)," Luo noted on X.
DeepSeek's V4 model also employs innovative techniques to minimize computing demands. With a context capacity of one million tokens, V4-Pro's KV cache is only 10% the size of its predecessor, allowing for a 98% reduction in costs compared to GPT-5.5 Pro while maintaining competitive performance.
U.S. Pricing Strategies and Market Dynamics
As U.S. firms like Anthropic and OpenAI raise their prices, the contrast with Chinese models becomes more pronounced. Claude Opus 4.7 charges $5 per million input tokens and $25 for output tokens, while GPT-5.5 has doubled its output pricing. In comparison, Chinese models like MiniMax M2.7 and Kimi K2.5 offer significantly lower rates, with MiniMax priced at $0.30 for input and $1.20 for output tokens.
This pricing disparity highlights a broader trend, with Chinese AI models consistently outperforming their U.S. counterparts in terms of cost-effectiveness. The recent adjustments further solidify this trend, particularly for workloads that benefit from frequent cache access, such as document processing and retrieval systems.
