Gemma 4 26B-A4B uncensored fine-tunes lag

The short answer is “yes, but not a great one yet.” There are already a few Heretic-based decensored builds, but this is still launch-week territory, so quality is uneven and the safest move is to let the dust settle before picking a fork.

• –Google shipped Gemma 4 26B with broad official tooling support, so the base model is the stable starting point
• –Early community variants like the Heretic-based decensored builds are showing up fast, but that speed comes with rough edges
• –One of the model-card notes already flags broken or repetitive behavior on some harmful prompts, which is a bad sign for production use
• –If you want fewer refusals today, prompt/system tuning on the official model may be more reliable than chasing the first random uncensored upload
• –Expect better options over the next 1-2 weeks as fine-tunes, quants, and inference support mature