METR Q4 2026 TH2 - AZsPCs Balto's Puzzle (Human Percentile): leaderboard

METR's Q4 2026 capability update (20 November 2026): where a model's best 48-hour run on the past Al Zimmermann's Programming Contest "Balto's Puzzle" ranks among that contest's 169 human entrants, who had about three months. 100 means beating every human entrant. One of the 15 contests in the TH2 suite.

Metric: Percentile Among Human Participants. Source: metr.org. Saturation forecast: Around February 2027. 10 models tracked. Updated 21 Nov 2026.

Top models

#ModelScoreOverall rank
1GPT-6 Astra99.41#10
2Claude Opus 5.595.27#9
3Claude Fable 5.195.27#20
4GPT-5.6 Sol78.11#37
5GPT-6 Luna73.37#105
6Claude Haiku 5.573.37#75
7GPT-5.469.82#135
8Claude Opus 4.755.62#85
9GPT-519.53#178
10Claude Opus 4.517.16#160

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-azspcs-balto-s-puzzle-human-percentile · How It Works · Data refreshed daily, snapshot 2026-11-21.