Local models are found able to handle 88.7% of chat queries

Claim

Stanford and Together AI proposed intelligence per watt as a metric, finding local models can already handle 88.7% of chat queries, while OpenAI quietly halved inference costs and Google's zero-shot tabular foundation model retired feature engineering behind a single SQL call.

Reported metric: 88.7% of queries local

Reporting date: 2026-07-01 (not necessarily the event date)

Sources

Attribution and reporting linked by this record; links do not imply independent verification.

• Welcome to July 1, 2026 (theinnermostloop.substack.com)