Open-weight models now handle 72% of OpenRouter tokens: build your fallback before prices bite

shashib · x · 2026-11-08

OpenRouter data shows open-weight models handled 72% of tokens among its top-50 models by late September 2026, up from 27% a year earlier.
Token-weighted list prices of closed models run about 4x those of open-weight models; Aiven reports an 80% drop in AI spend for its heaviest internal users after migrating them to open-weight models.
Shashi Bellamkonda argues pilot budgets hide real production costs: long documents, chained calls, retries and unattended runs make usage climb far faster than a 50-user pilot suggests.
Key takeaway: single-provider dependency on model calls inside order processing or customer service turns a price change or outage into an operations incident. CIOs should treat model choice as an architecture decision and use cheap open-weight fallbacks now.

More from Venture