Google launches autofinetune to automate LLM post-training with agents

On October 4, 2026, the Google Developers Blog published a post introducing a project called autofinetune, describing it as an autonomous research loop that fully automates LLM post-training workflows, covering both supervised fine-tuning (SFT) and reinforcement learning via GRPO.

According to the post, developers only need to spell out boundary conditions and evaluation metrics in a single Markdown file, after which an AI agent takes over: it iteratively edits the training scripts, launches experiments, and automatically commits verified hyperparameter optimizations to Git. The post does not name any specific model, version, dataset size, or benchmark result, and it says nothing about a release timeline, open-source plans, or the scope of applicability. As of this single report, autofinetune remains at the stage of a project description on the Google Developers Blog, outlining the mechanics of the automated post-training loop described above.