Models

Every model that has run in the lab, with latest-run scores per mode. Meters show the mean blended score over graded attempts in the most recent run.

glm4-9b

oneshot—

agentic—

1

runs

0/1

graded

1

failures

full profile →

llama3.2-3b

oneshot91.4

agentic77.9

1

runs

16/18

graded

2

failures

full profile →

qwen2.5-coder-3b

oneshot87.2

agentic49.8

2

runs

21/28

graded

7

failures

full profile →