METR Capability Index (MCI) - Q4 2026 Fit: leaderboard

METR's Q4 2026 capability update (20 November 2026): METR's new single ability score per model, from an item response fit over every task in TH1.1, TH2 and MirrorCode rather than a time-horizon curve. The scale is pinned at GPT-4o = 0 and Claude Opus 4.6 = 633, the number of days between their releases, so a point is roughly a day of frontier progress. Unbounded; higher is better. Includes four models listed only in the post's MCI-against-ECI comparison.

Metric: MCI. Source: metr.org. Saturation forecast: Around December 2027. 34 models tracked. Updated 21 Nov 2026.

Top models

#ModelScoreOverall rank
1GPT-6 Astra858.98#10
2Claude Opus 5.5854.1#9
3Claude Fable 5.1839.12#20
4Claude Opus 5796.23#26
5Claude Fable 5778.74#33
6GPT-5.6 Sol738.27#37
7Claude Opus 4.8708.6#65
8GPT-5.5705.69#54
9Claude Mythos Preview (Early Snapshot)671.22#44
10Claude Opus 4.6633#113
11Claude Opus 4.7631.23#85
12Claude Haiku 5.5627.8#75
13GPT-6 Luna614.6#105
14GPT-5.4593.18#135
15Gemini 3.1 Pro (Preview)568.91#99

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

Interactive version: theaggregate.ai/benchmark?slug=metr-capability-index-mci-q4-2026-fit · How It Works · Data refreshed daily, snapshot 2026-11-21.