Open-weight models' share of Databricks customers' tokens rose from 7.3% to 30.9% in four months

Databricks' same November 16 report compares the models enterprise customers used in July and in November (Databricks, 2026-11-16). Only 28% of Databricks customers still use a single model vendor, down from about 64% in July; most now connect to several.
Open-weight models' share of the tokens Databricks customers used rose from 7.3% in July to 30.9% in November. Over the same period, the share of companies using only one model vendor fell from about 64% to 28%. Three open-weight models also made the top ten on Databricks' own coding benchmark, with DeepSeek V4.1 and GLM 5.3 tied for third at a 93.1% success rate.
The open-weight share is much higher on developer platforms. Hugging Face announced the same day that its Inference Providers service now uses OpenRouter's routing to pick a provider (Hugging Face, 2026-11-16). Inference Providers lets users call many inference providers through one API. OpenRouter is a routing platform that connects to dozens of providers through a single API. Hugging Face cites OpenRouter's figures: open-weight models' share of tokens on OpenRouter rose from 26.8% in October 2025 to 71.7% at the end of September 2026.
The two sets of figures cover different users at different times, so neither stands in for the other. OpenRouter's users are mostly developers, and its figures run to the end of September. Databricks is an enterprise gateway, and its figures run to November.
With this change, Hugging Face hands the provider-picking layer to OpenRouter. Each call goes to a provider chosen on live price, speed and reliability, and a failed call is passed to the next provider within the same call.
Not proven yet: the OpenRouter figures appear only as cited by Hugging Face How we checked →
Next to watch: a second enterprise-side dataset, such as an enterprise AI spending survey, and whether the share of companies using a single vendor there also falls to around 30%. If it stays above half, the 28% may only reflect that Databricks customers already like to compare prices.
Source and checks
This item ran in the brief of Nov 17, 2026, alongside 13 other items. How we checked it, the quote checks and what was still missing that day are on that issue's How we checked page.
Cite this item
SecondSource (2026). "Open-weight models' share of Databricks customers' tokens rose from 7.3% to 30.9% in four months." https://secondsource.io/en/news/2026-11-17-2/. Accessed access date.