Frontier Log

Claude and GPT models on FrontierCode 2.0

Changed by Cognition on

Said it had changedPrinted a new figure

Cognition updated the cheating scanner on its FrontierCode 2.0 leaderboard to remove false positives on Nov 11, 2026, and rescored all results. At xhigh effort, Claude Fable 5.2 rose from 49.1% to 50.2%.

FrontierCode 2.0 Scores Changed After Cognition’s Rescore. FrontierCode 2.0: Claude Fable 5.2, xhigh 49.1% before and 50.2% after, Claude Opus 5.5, max 45.5% before and 47.3% after, GPT-6 Astra, max 42.8% before and 44.0% after, GPT-6.1 Sol, max 39.0% before and 39.6% after, Claude Fable 5.1, xhigh 32.4% before and 34.0% after.

Download image

At max effort, Claude Opus 5.5 rose from 45.5% to 47.3%. GPT-6 Astra rose from 42.8% to 44.0%, and GPT-6.1 Sol rose from 39.0% to 39.6%. At xhigh effort, Claude Fable 5.1 rose from 32.4% to 34.0%.

The earlier figures for those four models are from Frontier Log’s copy of the leaderboard of Oct 31, 2026. Cognition added Fable 5.2 on Nov 9, and its earlier figure is from Anthropic’s Claude Fable 5.2 post. That post prints the earlier figures for Opus 5.5 and Fable 5.1, and OpenAI’s GPT-6.2 post prints them for GPT-6 Astra and GPT-6.1 Sol.

Comparison

Claude Fable 5.2 (xhigh)

Before Anthropic’s Claude Fable 5.2 post, Nov 10 copy
Before: Anthropic’s Claude Fable 5.2 post, Nov 10 copy
After Cognition’s leaderboard, Nov 18 copy
After: Cognition’s leaderboard, Nov 18 copy

Claude Opus 5.5 (max), GPT-6 Astra (max), GPT-6.1 Sol (max), Claude Fable 5.1 (xhigh)

Before Cognition’s leaderboard, Oct 31 copy
Before: Cognition’s leaderboard, Oct 31 copy
After Cognition’s leaderboard, Nov 18 copy
After: Cognition’s leaderboard, Nov 18 copy

All benchmark corrections