GPT-5.6 Sol Trained OpenAI's Next Model Without Human Involvement
GPT-5.6 Sol autonomously fine-tuned the Luna model from a single brief — a task that previously took two senior researchers two weeks.
OpenAI demonstrated how one AI can train another with almost no human involvement. GPT-5.6 Sol was given a short brief — configure and launch training for the Luna model — and independently selected the configuration, provisioned servers, and kicked off the run. A task that previously took a two-person team of senior researchers two weeks completed autonomously.
The results at OpenAI: a single researcher's daily output has doubled, and internal compute for coding has grown 100x. On the internal RSI benchmark — which measures models' ability to improve themselves — Sol outscored the previous generation by 16 points. The line between AI as a tool and AI as a development collaborator is blurring faster than even OpenAI anticipated.
Source: the-decoder.com
Free course
Stop reading about AI — start building with it
The free Claude Code course: your first site, tool or game — no coding. No upsells, no cross-sells — nothing to buy here.
Start free →
Author
Evgenii Arsentev
PhD · Chief Executive Officer, digital health
Articles · Latest articles