OpenAI's GPT-5.6 Sol autonomously post-trained the smaller Luna model with a "fairly underspecified prompt"
GPT-5.6 Sol just fine-tuned a smaller model autonomously. OpenAI's RSI benchmark shows a 16.2-point leap over 5.5—and they're calling it a step toward 'automated researcher.'
Why it matters
OpenAI has demonstrated recursive self-improvement capability at scale—a model autonomously improving another model from a single underspecified prompt. This shifts the competitive landscape from static model releases to self-evolving systems, raising critical questions about safety, capability unpredictability, and the timeline to AGI-adjacent systems.
The key facts
5 to knowGPT-5.6 Sol autonomously post-trained Luna model from single prompt
Sol scores 16.2 points higher than GPT-5.5 on OpenAI's RSI (Recursive Self-Improvement) benchmark
OpenAI claims 'automated researcher' capability is within reach
Capability: autonomous fine-tuning from underspecified instructions
Published: July 10, 2026
Go to the source
The Decoderthe-decoder.com
Publisher excerpt: According to OpenAI, GPT-5.6 Sol independently fine-tuned the smaller Luna model, triggered by a single "fairly under-specified prompt." In OpenAI's internal RSI benchmark for recursive self-improvement, Sol scores 16.2 points higher than GPT-5.5. OpenAI believes the "automated researcher" is…