METR Q4 2026 TH2 - Budget NanoGPT: 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 speeds up a NanoGPT speedrun training script starting from record #78, with 48 hours on eight H100 GPUs; the score rises with the training time cut. Higher is better.

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

Top models

#ModelScoreOverall rank
1GPT-6 Astra0.1267#10
2Claude Fable 5.10.0824#20
3Claude Opus 5.50.0821#9
4Claude Opus 50.0552#26
5Claude Fable 50.0167#33
6Claude Opus 4.50.0121#160
7GPT-5.6 Sol0.0115#37
8Claude Opus 4.60.0091#113
9Claude Opus 4.70.0056#85
10Claude Opus 4.80.0043#65
11GPT-5.50.0042#54
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 · How It Works · Data refreshed daily, snapshot 2026-11-21.