Raymond James’ Token Observer report for August 2026 highlights a widening disparity in AI model token pricing, with the gap between the highest- and lowest-priced models expanding to roughly $50 per million tokens, up from $30 in July.
The expansion reflects elevated costs at the premium end of the market, driven by new cybersecurity-focused models. Average token pricing declined 25% month-over-month to $1.15, a figure Raymond James attributed to a concentration of lower-cost open-source models in the dataset, which may not fully represent broader industry conditions.
GPU pricing rose 1% in August, contrasting with an 8% increase reported by Semianalysis in July. Performance benchmarks across eight categories showed the top two models outperforming the bottom two by more than 300% in agentic and tool-use applications, with overall performance improving by approximately 30% across all categories.
Token volume data from OpenRouter indicated a 45% month-over-month increase in August, accelerating from roughly 40% gains in the prior three months. Raymond James cautioned that OpenRouter’s data likely skews toward developer activity and underrepresents enterprise and consumer usage.
Vercel’s August data on token spending and volume shares showed Hyper AI Labs—including Anthropic, OpenAI, Google, and Meta—holding 85% of token spending share and 46% of volume share. Chinese AI Labs, encompassing DeepSeek, Moonshot, and Z.ai, accounted for 13% of spending and 39% of volume.
