METR Q4 2026 TH2 - Budget NanoGPT (16h): 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 NanoGPT speedrun task (make a language-model training script faster, starting from record #78) in the variant METR's data labels 16h. Higher is better.

Metric: Mean Score (0-1). Source: metr.org. Saturation forecast: Around October 2028. 15 models tracked. Updated 21 Nov 2026.

Top models

#ModelScoreOverall rank
1Claude Opus 50.102#26
2Claude Opus 5.50.0666#9
3Claude Fable 5.10.0633#20
4Claude Fable 50.0448#33
5GPT-6 Astra0.0419#10
6GPT-5.6 Sol0.0126#37
7GPT-5.50.01#54
8Claude Opus 4.70.0046#85
9Claude Opus 4.60.0034#113
10Claude Opus 4.50.0033#160
11Claude Opus 4.80.0009#65
12GPT-4o0#751
13GPT-50#178
14GPT-5.40#135
15O10#403

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-budget-nanogpt-16h · How It Works · Data refreshed daily, snapshot 2026-11-21.