Terminal-Bench 4.0 leaderboard: Claude Sonnet 5.5 tops at 70.6%
NielsRogge · x · 2026-11-18
Terminal-Bench 4.0 ships with calibrated task time/CPU/memory limits, 19 fixed tasks, 8 removed saturated tasks, and 66 tasks with an 8-hour agent timeout. Leaderboard top 10: Claude Sonnet 5.5 (70.6%), Claude Mythos 5.2 max (70.5%), Claude Opus 5.5 (66.36%), Kimi K3.1 Pro (65.7%), Gemini 4 Carbon (65.2%), Claude Mythos 5.1 max (60.9%), GPT-6 Astra (57.9%), Gemini 4 Argon (57.4%), Muse Wonder Full (57.1%), HERMES GPT-5.6 Sol (55.5%).
More from Models
- SenseNova-Loop-9B, a looped-transformer multimodal model, trends on Hugging Face — sensenova · 2026-11-18
- Third-party test suggests Claude was trained on protein sequences at scale — AllThingsApx · 2026-11-18
- Power users brace for December's shared ChatGPT and Codex usage limits — F1_average_enjoyer · 2026-11-18
- Pre-training trend watch: dropping LR decay for constant LR plus checkpoint averaging — rasbt · 2026-11-18
- Mistral's Le Chonk crowned best model name, free Vibe tier for Europe praised — TheTuringPost · 2026-11-17
- Dev exhausted 128 Claude accounts to dodge usage limits, takes week off — thomasahle · 2026-11-17