For the first time, open-weight models have processed more than half of the tokens passing through a commercial AI gateway in production. This is the central conclusion of the latest "AI Gateway Production Index" published by Vercel, which aggregates anonymized traffic data from its AI Gateway for August 2026. The report shows a significant shift in how businesses and developers are using artificial intelligence models, with open-source models gaining substantial ground over proprietary ones.
According to Vercel's report, open models represented only 7% of the tokens processed in December 2025. Eight months later, in August 2026, they captured 56%, surpassing all proprietary models combined. This growth has been continuous since April. Despite this increase in usage, open-weight models account for only 14% of total spending on the platform, indicating that closed laboratories still hold the majority of value, even as they lose volume in terms of token processing.
The rise in the use of open-weight models has had mechanical effects on prices: the average cost per token declined by 23.2% in August, a third consecutive monthly decline and the largest since April. Among teams processing more than ten million tokens over two consecutive months, the median cost per token fell by 7.6%, more than double the decline observed in July (2.9%). This trend shows that as more users turn to open models, the cost per token is decreasing, likely due to increased competition and efficiency.
The report documents a rapid client reallocation within the Anthropic range. Fable 5, the most powerful model from the laboratory, saw its share of spending rise to 13.2% in July after the lifting of restrictions on American exports that had suspended its access. A month later, this share fell to 4.9%, while Opus 5, sold at about half the price, saw its share jump to 22.5%. Nine out of ten teams that used Fable reduced their consumption, with the majority switching to Opus 5 rather than a competitor. Despite this shift, Anthropic's dominant position remains intact, with its models capturing at least 61% of the gateway's total spending each month since December and 64% in August.
Open-Weight Models Surpass Proprietary Models in AI Gateway Traffic
AI-rewritten from original reportingHow it works
ai-gatewayopen-weight-modelstoken-volumecost-efficiencymodel-adoptionanthropic



