(Yicai) Aug. 7 -- Leading artificial intelligence companies have been repricing their offerings recently, as the low-cost open-source approach of Chinese rivals continues to disrupt the industry.

US-based OpenAI, creators of ChatGPT, announced yesterday that Free and Go users will get unlimited text chats with the latest version, GPT-5.6 Luna, which will directly benefit about one billion users worldwide.

On the same day, Hangzhou-based DeepSeek announced plans to raise the prices of its application programming interface services in the near future, warning of a “significant” increase.

OpenAI announced an 80 percent price cut for GPT-5.6 Luna on July 30, with the output price reduced from USD6 per million tokens to USD1.20, and the input price lowered from USD1 to 20 US cents. On the same day, the official API version of DeepSeek V4 Flash was launched.

The price cuts failed to prevent a further loss of developers for OpenAI, which has been feeling the heat from lower-priced Chinese competitors. Like most leading American AI firms, OpenAI’s products are proprietary, while the open-source models from Chinese firms are nearly as effective and far cheaper.

According to data from third-party model aggregation platform OpenRouter, the call volume of DeepSeek V4 Flash’s old and new editions totaled 11.31 trillion tokens this week (Aug. 3 to today), while that of GPT-5.6 Luna amounted to 3.54 trillion tokens, a difference of nearly three times.

It should be noted that OpenRouter’s data reflects the number of calls distributed through the platform, and does not include the direct calls made by developers from OpenAI. Even so, the gap still indicates that price reductions have not effectively curbed the trend of developers migrating to Chinese open-source models.

The logic of developers “voting with their feet” is not complicated. According to a report by Huatai Securities, the package price of DeepSeek V4 Flash is about 6 US cents per million tokens, about 65 percent lower than that of OpenAI’s GPT-5.6 Luna, and the single task cost is about 57 percent lower. If the performance gap is not large, the choice for developers is simple.

As the previous price reduction was ineffective, OpenAI has turned to a more aggressive free strategy. GPT-5.6 Luna is 62 percent lower in factual error rate, it said, than the default free model, GPT-5.5 Instant. Among paid versions, GPT-5.6 Sol is the main revenue source, and it now powers both instant and deep reasoning for Plus and Pro users, Open AI also announced yesterday.

DeepSeekV4 Flash quickly topped the third-party global call volume list, but with the widespread use of AI agents, the computing power needed for a single task can be 100 times that of a regular dialog task, and the surge in demand for computing power during peak hours leads directly to a decline in service stability. 

This means the low-price strategy has gradually evolved into a dilemma, as the more calls there are, the more losses there will be for the company.

According to statistics from Kaiyuan Securities, China’s Zhipu AI raised the prices of its API services by 83 percent in the first quarter, and the call volume surged 400 percent. Alibaba Cloud, Tencent Cloud, and Baidu AI Cloud also raised the prices of AI-related products and services earlier this year. 

DeepSeek’s move to change its pricing strategy reflects the strategic shift among Chinese AI firms from offering low prices in order to build market share to pricing based on value.

Editor: Tom Litting