In the news
The best mid-size model to fine-tune was the third-best model to prompt
Distil Labs · Published · 3 min read
In 30 seconds
- What happened
- Nemotron 3.5 Lightning ranked third in zero-shot performance but first after fine-tuning across six tasks, beating a 550B teacher model on four tasks.
- Why it matters
- Engineers choosing which mid-size model to fine-tune for production features should prioritize fine-tuning capability over base model zero-shot rankings.
- Watch out
- Results are tight with many ties; text2sql performance at 0.850 is weak; small test sets mean sampling noise significantly affects rankings and conclusions.
Listen to this summary
- prompt
- fine-tun
The patterns behind this
Each one covers how the technique works, when it earns its cost, and where it breaks.
The Agent Architect
One pattern, one tradeoff, one production failure story. A short weekly briefing for people building agentic systems.
Weekly email, one-click unsubscribe. We only use your address to send the briefing.