AI token prices hit annual low as DeepSeek and cheap Chinese models pressure the market

[Lantern]

Average prices for AI inference have fallen to their lowest level of the year so far, driven by a wave of low cost open weight Chinese models alongside a genuine increased focus on price efficiency from US developers, according to research from investment bank Jefferies

Inference prices per million tokens ranged between 1.16 and 1.18 dollars from August 6th to 8th, the lowest level recorded so far this year, significantly down from a 2.04 dollar average on May 31st and a 1.45 dollar average in late July, citing analysis from Silicon Data, the downward trend follows the recent release of DeepSeek's V4-Flash-0731 model, which costs just 0.03 dollars per task, a figure Jefferies described as materially lower than both domestic and international competitors

That specific model has rapidly climbed to the top of global token consumption rankings on OpenRouter, the platform that lets developers toggle between different AI providers, accounting for 27 percent of total processing volume on the platform as of Monday, ahead of Google at 25 percent, DeepSeek's model was the most used on the platform last week overall, handling 8.22 trillion tokens, followed by Tencent's Hy3 model at 7.13 trillion tokens and an earlier April version of V4-Flash at 6.05 trillion tokens

The pricing pressure isn't purely a Chinese phenomenon either, Jefferies analysts noted the downward trend coincided with an increasing emphasis on cost efficiencies in both the US and China simultaneously, last month OpenAI slashed prices for its latest GPT-5.6 model series by up to 80 percent, while Anthropic's Claude Opus 5 reportedly delivers performance comparable to the flagship Fable 5 model at half the price

Despite falling per token prices, overall AI costs at enterprises have nevertheless spiralled, largely because of the increasing use of agentic tools capable of working independently on tasks for hours or even days at a stretch, these tools can consume large and genuinely unpredictable numbers of tokens, and according to recent research from global consulting firm EY, businesses largely have no reliable way to monitor expenses of this kind, meaning falling headline token prices haven't necessarily translated into falling total AI spending for a lot of companies

AI Token Prices Fall To Annual Low | Silicon UK Tech Newswww.silicon.co.uk