How-ToAI ModelsNovember 16, 2026

Databricks distills gpt-oss-120b into 0.8B Qwen3.5 model

Read original source →youtube.com

A 0.8B Qwen3.5 student fine-tuned with LoRA on AI Runtime matched OpenAI's gpt-oss-120b on 100 held-out support tickets (95 vs 95) at roughly a third of the latency. Labels came from ai_query calls to the 120B teacher.

1 source

More stories today

Open the live feed