Terminal-Bench 2.0 on Gemini 3 Pro Preview

89 tasks, 5 attempts, open network. Updated 25 Oct 2026.

Pass rate by harness, higher is better

Reference

Δ vs sageagent, pts

sageagent65.2%95% interval 56.0–73.5sageagent is the reference.

Ante64.7%95% interval 56.2–72.1Ante minus sageagent: −0.4 points, 95% interval −6.7 to +5.4, over 89 tasks both ran; too close to call.

Every harness on this board, in its order
Harness	Rank	pass^5	Last published
sageagent	1–2	53.9%	25 Oct 2026run #79
Ante	1–2	44.9%	25 Oct 2026run #43