List your company
ResearchFormerly WizWand

Terminal interaction agent task on Terminal-Bench (dev)

37.4TB-dev ScoreGPT-5 Nano
4.43212.99121.5530.109Jul 6, 2026
Updated 3mo ago

Evaluation results

MethodLinks
GPT-5 NanoSource=Openthoughts, Context=Baselines2026.0737.4
GLM-4.6Context=Baselines2026.0729.9
Qwen3-235B-A22BSource=Openthoughts, Context=Baselines2026.0726.9
OP-SHORTBase Model=Qwen3-8B, Teacher Model=GLM-4.6, Prefix Training Data=10%, BC Data=5%, On-policy Data=5% (K = 3), RL=02026.0719.1
BC+OPBase Model=Qwen3-8B, Teacher Model=GLM-4.6, Type Detail=different tasks, Prefix Training Data=10%, BC Data=15%, On-policy Data=10% (K = 3), RL=720 tasks2026.0718.3
OP-SHORTBase Model=Qwen3-8B, Teacher Model=GLM-4.6, Prefix Training Data=10%, BC Data=2%, On-policy Data=2% (K = ∞), RL=02026.0718
Pure BCBase Model=Qwen3-8B, Teacher Model=GLM-4.6, Prefix Training Data=100%, BC Data=100%, On-policy Data=0, RL=720 tasks, Source=Openthoughts2026.0717.3
Pure BCBase Model=Qwen3-8B, Teacher Model=GLM-4.6, Prefix Training Data=100%, BC Data=100%, On-policy Data=0, RL=720 tasks2026.0716.7
Pure OPBase Model=Qwen3-8B, Teacher Model=GLM-4.6, Prefix Training Data=10%, BC Data=0%, On-policy Data=10% (K = ∞), RL=02026.0716.7
OP-SHORTBase Model=Qwen3-8B, Teacher Model=GLM-4.6, Prefix Training Data=10%, BC Data=2%, On-policy Data=2% (K = 3), RL=02026.0716.1
BC+OPBase Model=Qwen3-8B, Teacher Model=GLM-4.6, Type Detail=different tasks, Prefix Training Data=10%, BC Data=15%, On-policy Data=10% (K = 3), RL=02026.0715.1
Pure OPBase Model=Qwen3-8B, Teacher Model=GLM-4.6, Prefix Training Data=10%, BC Data=0%, On-policy Data=10% (K = 3), RL=02026.0714.7
Pure BCBase Model=Qwen3-8B, Teacher Model=GLM-4.6, Prefix Training Data=75%, BC Data=75%, On-policy Data=0, RL=02026.0714.2
Pure BCBase Model=Qwen3-8B, Teacher Model=GLM-4.6, Prefix Training Data=100%, BC Data=100%, On-policy Data=0, RL=02026.0713.8
Pure BCBase Model=Qwen3-8B, Teacher Model=GLM-4.6, Prefix Training Data=10%, BC Data=10%, On-policy Data=0, RL=02026.0711.9
Pure BCBase Model=Qwen3-8B, Teacher Model=GLM-4.6, Prefix Training Data=50%, BC Data=50%, On-policy Data=0, RL=02026.0711.6
Pure BCBase Model=Qwen3-8B, Teacher Model=GLM-4.6, Prefix Training Data=25%, BC Data=25%, On-policy Data=0, RL=02026.0710.7
Qwen3-8BSource=Openthoughts, Context=Baselines2026.075.7