AI token prices fall to record lows amid the price war between models

A key gauge of artificial intelligence prices has hit a fresh all-time low, in the latest sign that the cost of using large language models is collapsing in an increasingly competitive market. The LLM Token Expenditure Index, compiled by market intelligence firm Silicon Data to track the daily benchmark price of a large language model token, fell to 97 cents on Monday, its lowest reading since it was created late last year and a drop of more than half from the peak recorded early this summer.

A decline that steep cuts both ways. For users of chatbots such as OpenAI's ChatGPT, Anthropic's Claude or Google's Gemini, it means paying less to run their queries. But for the companies operating those models it is bad news: a lower price index gets consumers used to expecting reduced rates, eroding providers' pricing power.

As Charles-Henry Monchau, chief investment officer at Syz Group, explained on Tuesday, much of the recent drop reflects the rise of Chinese open-source models such as Moonshot's Kimi K3, which can offer lower prices than the alternatives from the big frontier model labs. On top of that, OpenAI announced price cuts in late July for two of its GPT-5.6 models, while other labs have launched offerings with "dynamic pricing" capabilities, allowing access rates to rise or fall with demand. All of it adds downward pressure on the market price of tokens.

"Frontier model labs are the most directly exposed," Monchau wrote. "Token deflation compresses the revenue line while compute commitments remain fixed. The strategic response is already visible: the competitive moat has to shift from raw model capability — where the gap with open-weight models is now measured in months — toward distribution, memory and context," he added. Monchau also noted that the industry-wide decline in the cost of producing a token has contributed to the lower prices reflected in the index.

The index's deterioration comes at a delicate moment, as both OpenAI and Anthropic filed confidential IPO paperwork with regulators this summer. Falling token prices could add fresh pressure on their margins just as both companies weigh when, and whether, to go public.

Cheaper tokens are also forcing investors to rethink their forecasts for the return on capital invested in building AI infrastructure, into which tech giants such as Nvidia and Microsoft have poured billions of dollars to expand capacity. According to Steve Hou, head of research at Silicon Data, the recent drop could indicate that, between frontier models and their cheaper competitors, there is already enough supply to "provide sufficient capabilities for most tasks."

The stock market itself reflected some unease on Tuesday: tech stocks led the declines, with the Nasdaq Composite falling nearly 1% and the S&P 500 slipping 0.4%.

Taken together, the news points to a phase of maturity — and mounting competitive pressure — in the language model business: what until recently was a race centered on raw model capability is beginning to turn, according to the analysts cited, into a fight over distribution, memory and context, as the price of a token ceases to be a sustainable competitive advantage.