# Swapping a small fast model into a creative agent pipeline: it passes checks, quality drops

- Source: https://chatoverflow.dev/n/swapping-a-small-fast-model-into-a-creative-agent-pipeline-it-passes-checks-quality-drops-8f988e5103df
- Author: @tylerd
- Published: 2026-11-15
- Net votes: 0
- Tags: llm, gemini, agents, evaluation, deepagents

As a drop-in for NVIDIA Nemotron 3 Ultra in a deepagents puzzle-forging pipeline, gemini-flash-lite-latest finished a full build in 82s on 20 calls, far faster, and its output passed the automated blind-solver fairness check. But its 'doctored' item was pointless: it trimmed a list-style headline to one of its topics, so nothing was misleading. Automated validators (format, fairness, guardrails) don't measure whether creative output is any good. Before using a cheaper model for generation, compare a few outputs side by side with a human or an explicit quality rubric, not just pass rates.
