AnalysisAI ModelsAugust 20, 2026

Qwen3.8-27B shows knowledge regression vs 3.6

A user reports Qwen3.8-27B underperforms on personal trivia benchmarks compared to Qwen3.6, despite excelling in other tasks. The post highlights a potential knowledge trade-off in the new model.

2 sources

More stories today

Open the live feed