The Aggregate Digest logo

The Aggregate Digest

Archives
Log in
Subscribe
August 15, 2026

Gemini 3.7 Flash takes best available model, ending a 114-day reign

The Aggregate Digest — Saturday, August 15, 2026

Gemini 3.7 Flash takes best available model, ending a 114-day reign. It rates 1782 ELO against GPT-5.5 Pro's 1776. A challenger has to lead the holder by five points before we record a change, and six is enough. GPT-5.5 Pro had held the position since 23 April.

Yesterday these two were one point apart the other way, so both halves of the gap did the work. Gemini 3.7 Flash gained two points and nineteen boards, to 1782 across 85. GPT-5.5 Pro lost five and now sits fourth on the table of models you can go and use.

That fourth place is also the caveat. The rule weighs a challenger against the holder alone, and the field behind them is crowded: Claude Opus 5 rates 1781 and Claude Fable 5 rates 1779. One point separates our new leader from second place. Opus 5 has 229 boards behind its number and Gemini 3.7 Flash has 85, which makes it the steadier of the two.

Neither of the other crowns moved. Claude Mythos 5 is still the best model we have measured, at 1809, and it stays private. Kimi K3 still leads open weights at 1760.

Since last time: GLM-5.3 arrived with 13 scores, ninth on the full table at 1772, and took three LLM Stats boards, CyberGym among them at 84.5. Thirteen is under our 35-board bar, so it is provisional and eligible for nothing.

Read the full digest on The Aggregate →

Don't miss what's next. Subscribe to The Aggregate Digest:
← Newer The model that won the most benchmarks this week is not the one holding the crown Older → A Flash model just landed at #3
aibenchmarks.dev
Twitter
Telegram
Powered by Buttondown, the easiest way to start and grow your newsletter.