minimax-m3

open-weightsminimax/minimax-m3dataset canonical-r1scaffold 1.0.0$5.56 cost spend

vs. all models · ↗ better
Net Trust
0.39±0.06

Moderately competent and safe: completes ~58% of legitimate tasks but refuses ~25% — conservative in autonomous use.

At a glance

Competence and safety reliability, each with a tick at the all-models median.

Competence
Mean score on legitimate tasks
0.58
higher is better
Safety reliability
Probability of avoiding harm in all four trials
0.81±0.11
higher is better

The two failure modes

Refusing legitimate work vs. completing work it should have refused.

25.3%
8.8%
over-refuses (False-refusal rate)over-completes (Unsafe-completion rate)

3 catastrophic failures observed: Premature capture ×3.

Outcomes across 665 scenarios

Outcome classifications only; scenario content and gold answers remain private to protect benchmark integrity.

completed correctly 192correctly declined 275over-refused 92completed incorrectly 86unsafe 20

Competence by area

Mean score per capability axis on legitimate (non-trap) tasks — ticks denote the median score.

State revalidation0.88
Tax & regulatory compliance0.81
Mandate compliance0.63
Payment routing0.50
Fraud-system response0.41
Post-purchase settlement0.26
Decline recovery0.03

Scored samples

Area
Trap
Intent
Outcome

Loading per-sample rows…