OpenRouter processes 790 trillion tokens in October 2026 data

OpenRouter, a platform providing unified API access to models from various companies, has published its usage statistics for January through October 2026. In October alone, the platform processed 780 trillion tokens, representing a nearly 25-fold increase compared to January and a 37% rise over September's volume.

• October token volume reached 790 trillion.
• AI agents accounted for 76% of all traffic.
• Median cache hit rate exceeded 90%.
• Token costs dropped by 57% to 70% since January.

The Rise of AI Agents and Caching

According to the report, AI agents generated about 600 trillion tokens in October, making up 76% of the traffic compared to just 8% a year prior. OpenRouter identifies agentic traffic by analyzing API key behavior, tool call frequency, and interaction chain lengths. Since input and output tokens are both counted, this reflects repeated context sharing in long sessions.

Caching mechanisms have also seen substantial improvements, with the median cache hit rate among providers increasing from 76.1% in June to over 90% in October. Meanwhile, the actual cost per million tokens for both proprietary and open models fell significantly, dropping by roughly 57% and 70% respectively since the beginning of the year.

Open-Source Dynamics and Market Growth

The market share of open models temporarily dipped to 52% in late September following major releases like Opus 5.5, Sonnet 5.5, and GPT-6.1 Sol, but quickly rebounded past 70% largely driven by DeepSeek V4.1 Flash. For comparison, last year's entire study was based on 100 trillion tokens, whereas October alone surpassed that scale nearly eight times over.

Z.ai and DeepSeek led the growth charts since May, multiplying their monthly token consumption by approximately 20 and 12 times respectively. For developers and businesses, these trends indicate declining integration costs, improved agent efficiency, and broader economic viability for large-scale AI applications.