Aug. 22 at 12:02 AM
Raymond James’ August 2026 Token Observer report found that the price gap between the three most expensive and three cheapest AI models widened to about
$50 per million tokens, from
$30 in July, driven by new cybersecurity models at the high end. Silicon Data reported a 25% MoM drop in average token pricing to
$1.15, likely reflecting its heavy concentration of low-cost open-source models.
AI model performance improved about 30% across eight benchmarks, with the top two models outperforming the bottom two by more than 300% in agentic and tool-use applications. OpenRouter token volumes surged 45% MoM in August, accelerating from roughly 40% monthly gains over the prior three months, although its data likely underrepresents enterprise and consumer usage.
Vercel data showed Hyper AI Labs—including Anthropic, OpenAI, Google and Meta—accounted for 85% of token spending versus 13% for Chinese AI labs, while volume share was more balanced at 46% and 39%, respectively.
$GOOGL $META