List your company

Software Engineering Reasoning on Terminal-Bench (TB2) 2.0

36Resolution RateTACT-GATE
14.88820.36925.8531.331May 7, 2026

Evaluation results

MethodLinks
TACT-GATEBackbone=Qwen3.5-27B2026.053641.2
TACT-ADDBackbone=Qwen3.5-27B2026.0535.942.7
AGENTDEBUGBackbone=Qwen3.5-27B, rollouts=at least two per task2026.0532.881.7
TACT-CAPBackbone=Qwen3.5-27B2026.0532.636.5
AGENTREFLECTBackbone=Qwen3.5-27B2026.053246.8
RECAPBackbone=Qwen3.5-27B2026.0531.149.5
Base ModelBackbone=Qwen3.5-27B2026.0530.348.2
REMINDERBackbone=Qwen3.5-27B2026.0529.850.4
TACT-GATEBackbone=Gemma-4-26B-A4B-it2026.0519.116
TACT-ADDBackbone=Gemma-4-26B-A4B-it2026.0518.916.5
AGENTDEBUGBackbone=Gemma-4-26B-A4B-it, rollouts=at least two per task2026.0517.231.5
TACT-CAPBackbone=Gemma-4-26B-A4B-it2026.051713.9
RECAPBackbone=Gemma-4-26B-A4B-it2026.0516.718.4
AGENTREFLECTBackbone=Gemma-4-26B-A4B-it2026.051617.7
REMINDERBackbone=Gemma-4-26B-A4B-it2026.0515.918.6
Base ModelBackbone=Gemma-4-26B-A4B-it2026.0515.718.2