Skip to content
mager-bench1.3
16 attempts5 models
Low reasoningsame setting
4,096 tokensprompted target
SubscriptionCodex CLI

The calibration log

Ordered by start time. The score is exposed faults out of eight.

Latest run protocol

The September 30 calibration resumed after an interruption; its original protocol retains that execution note. The October 1 protocol records the new subjects and availability checks. No missing output is assigned a score.

One frozen suite.

All displayed results are regraded by the exporter before publication. The suite fingerprint must match.

a56a4aadf5b20a319c7a57309608477d9e523da3d99c37a69bb6353f6b16f0c1

Get the JSON