GLM-5.3 Flash

Qwen3.8 Scores 43% on Local Coding Tasks, Closing the Long-Tail Gap

Qwen3.8 27B solved 43% of 202 real GitHub tasks across 13 niche languages, up from 17% for Qwen3.6.