The price of tokens, the basic units that generative artificial intelligence (AI) models use to process information and generate answers, has fallen to a record low. The drop appears to stem from intensifying price-cut competition among major AI corporations amid low-priced AI from China.
CNBC reported on the 1st that the "large language model (LLM) token expenditure index," an AI token price gauge compiled by market research firm SiliconData, stood at $0.97 on the 31st of last month. That is the lowest level since the index was created late last year. Compared with the late-May peak this year ($2.05), it has been cut in half. The index is calculated by tracking the prices at which tokens—the units LLMs use to process text—are actually traded in the market.
The index has generally trended downward since June, and the decline has grown more pronounced this week.
The drop in token prices is negative for the revenue structures of model developers such as OpenAI and Anthropic and for the outlook on recouping large-scale AI investments. While lower token prices reduce the expense burden for consumers using AI chatbots such as ChatGPT, Claude and Gemini, CNBC noted that profitability could worsen for model developers supplying them, including OpenAI, Anthropic and Google.
Analysts say China's low-priced open-weight (publicly released weights) models have recently led the decline in token prices. Charles-Henri Monchau, chief investment officer (CIO) at Syz Group, said, "Models like Moonshot AI's 'Kimi K3' are offering lower prices than models from leading U.S. AI corporations, pulling down prices across the market."
Earlier, at the end of July, OpenAI cut prices for two models in the GPT-5.6 lineup, and other companies adopted "dynamic pricing," which allows usage prices to be adjusted to demand, a development seen as adding to the downward pressure on prices.
Monchau, the CIO, said, "AI corporations' fixed costs related to computing (compute resources) remain in place, but falling token prices are hitting sales," adding, "As the performance gap between open-weight models and top-tier closed models narrows by the month, the industry needs to move away from strategies reliant on performance advantages and focus on distribution channels, memory, and context processing capabilities."
In particular, the drop in token prices is a headwind for OpenAI and Anthropic, which are preparing for initial public offerings (IPOs). The two companies recently submitted confidential listing applications to U.S. regulators. The market has begun to question whether the AI industry's large-scale investments can deliver the expected revenue.