API
< Back to situations

Monitor this situation.

[SITUATION] · [QUIET] · [TECHNOLOGY]

2 clusters · 8 sources · 16 days · First seen · Last updated

AI model usage shift toward low-cost models

Overview

Data indicates a significant shift in AI usage patterns toward lower-cost, non-frontier, and open-weight models. In August, open-weight models processed 56% of all tokens via Vercel’s AI Gateway, a substantial increase from 13% in April, despite representing only 14% of total spending. This trend aligns with a continuous decline in average price per token.

Further analysis of the OpenRouter platform shows that during the week of August 10, 84% of token volume was directed toward models not considered state-of-the-art. A small group of six models accounted for roughly 80% of volume, providing approximately 77% of frontier model performance at only 2.5% of the cost. Additionally, Chinese models have seen increased adoption, accounting for over 30% of US company token consumption on OpenRouter, up from approximately 4.5% in early 2025.

Entities

Anthropic · Google · OpenAI · LMCache · Everpure

Timeline

  1. [TECHNOLOGY] 2 sources
    OpenRouter users shift 84% of token volume to non-frontier AI models

    OpenRouter data shows 84% of token volume is now driven by non-frontier AI models, as developers prioritize cost-effective, “good enough” performance over expensive state-of-the-art models.

  2. [TECHNOLOGY] 6 sources
    Open weight models process majority of AI tokens

    Open weight AI models processed 56% of all tokens via Vercel’s AI Gateway in August, surpassing closed weight models for the first time as per the AI Gateway Production Index.

Sources

bcdvideo.com · blog.purestorage.com · cryptobriefing.com · key4biz.it · lifehacker.com · lifehacker.com.au · techcentral.ie · tokenpost.com