Team finetunes Gemma 12B for audio proofreading, benchmarks it against Gemini

ojasvi_yadav · x · 2026-09-04

Developer averma12 shares a hands-on comparison: the team finetuned Gemma 12B for audio reasoning in an AI proofreader for audio transcripts, then ran it against their benchmark against the current SOTA Gemini harness (gemini-flash-3.5, optimized for cost). Numbers are shown in an attached chart. A useful reference case for finetuning small open models to replace frontier APIs on vertical tasks.

Original post →

More from Models

Models channel →