METR Q4 2026 TH2 - Judge Run (Newer Models' Tasks): leaderboard

METR's Q4 2026 capability update (20 November 2026): mean score from 0 to 1 on one task of TH2, METR's new long-task suite. The agent grades another agent's attempt at building an evaluation task, using a rubric, and is scored on agreement with METR's human grader. This half uses tasks built by models released from 9 June 2026, where every judge agrees with humans less. Higher is better.

Metric: Mean Score (0-1). Source: metr.org. Saturation forecast: Around May 2027. 15 models tracked. Updated 21 Nov 2026.

Top models

#ModelScoreOverall rank
1Claude Opus 5.50.4586#9
2Claude Fable 5.10.35#20
3GPT-6 Astra0.2653#10
4Claude Opus 50.1769#26
5Claude Opus 4.70.1476#85
6Claude Fable 50.1461#33
7Claude Opus 4.50.1095#160
8Claude Opus 4.60.0919#113
9GPT-5.6 Sol0.0666#37
10GPT-5.40.0551#135
11GPT-50.0387#178
12O10.0367#403
13GPT-5.50.0288#54
14Claude Opus 4.80.0263#65
15GPT-4o0#751

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=metr-q4-2026-th2-judge-run-newer-models-tasks · How It Works · Data refreshed daily, snapshot 2026-11-21.