AnalysisAI AgentsJuly 10, 2026

Untuned 27B model beats tuned 75B model on agentic tasks

An untuned 27B model completed all agentic tasks on a neutral system prompt in 6-9 tool calls, while a tuned 75B model required a hand-tuned profile and used twice as many turns. The 27B outperformed the 75B despite being smaller and without fine-tuning.

1 source