Atlas

Benchmarks

Epoch Automation Reports (November 2026)

Human rubric evaluation of six native agent systems on seven tasks in four categories, from Epoch's November 17, 2026 report. That report replaced the October 8 task suite (six tasks and the data-explorer category removed; Token prices, Chess variants and Hidden messages added), reworked the prompts and moved grading to a 1-4 scale per criterion, so it is kept apart from the October release. A task scores its rubric points over the 4-point maximum, weighted by section, so the lowest possible score is 25; a category weights its easy, average and difficult tasks 2:3:2; the overall is the unweighted mean of the four category scores. Highest available reasoning was selected, but exact native effort values were not disclosed; memory and skill writing were disabled.

Models
6
Top score
81.7
Leader
Claude Fable 5.2
Updated
Nov 17, 2026
Resolution
—
Saturation
—

No signal or difficulty estimate. Too few models have been measured on this benchmark to estimate its difficulty.

Organizations
6 dated models · 3 frontier modelsClosed weightsOpen weights
Data: Epoch AIatlas.village.dev

Epoch Automation Reports (November 2026) leaderboard

#ModelSettingScoreSource
1Claude Fable 5.2thinking · Epoch Automation Reports: Claude Code; highest available reasoning setting (native value unreported); memory and skill writing disabled
81.7
2Claude Opus 5.5thinking · Epoch Automation Reports: Claude Code; highest available reasoning setting (native value unreported); memory and skill writing disabled
77.8
3Gemini 4 Argonthinking · Epoch Automation Reports: Antigravity; highest available reasoning setting (native value unreported); memory and skill writing disabled
67.2
4GPT-6.2 Astrathinking · Epoch Automation Reports: Codex; highest available reasoning setting (native value unreported); memory and skill writing disabled
64.9
5Claude Fable 5.1thinking · Epoch Automation Reports: Claude Code; highest available reasoning setting (native value unreported); memory and skill writing disabled
64.4
6GPT-6 Astrathinking · Epoch Automation Reports: Codex; highest available reasoning setting (native value unreported); memory and skill writing disabled
55.5
Loading Atlas data…