Raymond James sees token pricing gap widen in August 2026

Raymond James observed a widening gap in token pricing for AI models in August 2026, with the difference between the top and bottom three models reaching around $50 per million tokens. This increase from $30 in July was attributed to the emergence of new cybersecurity-focused models. Silicon Data reported a 25% month-over-month decline in average token pricing to $1.15, likely due to the dominance of open-source models.

The report also highlighted that the top two AI models outperformed the bottom two by more than 300% in eight benchmark categories, including general knowledge, agentic and tool use, coding, cybersecurity, and health and legal applications. Performance improvements across all categories amounted to approximately 30%. August saw a 45% month-over-month surge in token volume, driven primarily by developers, according to OpenRouter data, which may not fully represent enterprise and consumer usage.

Vercel data indicated that Hyper AI Labs, comprising Anthropic, OpenAI, Google, and Meta, secured 85% of token spending share, while Chinese AI Labs accounted for 13%. Volume share was more evenly distributed, with Hyper AI Labs at 46% and Chinese AI Labs at 39%. GPU pricing rose by 8% month-over-month, while Silicon Data reported a 1% increase in August.