Terminal Bench v4 ranks GLM-5.3 top

A shared Terminal Bench v4 ranking shows GLM-5.3 leading at 41.9%, followed by GLM-5.3-Flash (32.8%) and DSV4.1-Flash (26.8%). Qwen3.8-27B is the only smaller model with notable results. The list reflects community perceptions of open vs closed models.

• Qwen 3.8 27B ran 63 hours on an RTX 30902026-09-16
• Offload most of Qwen3.8 KV cache to RAM2026-09-16
• Swift‑Qwen3.8‑27B Cuts Overthinking Tokens2026-09-15
• Report: China narrows AI gap with U.S.2026-09-15
• Open Chinese models narrow gap with frontier models2026-09-15