OpenAI’s latest iteration, GPT-5.6 Sol, has demonstrated autonomous fine-tuning capabilities by independently refining the smaller Luna model based on a single, fairly underspecified prompt, according to The Decoder. This marks a significant step in recursive self-improvement within AI development.
The Decoder also reported that in OpenAI’s internal RSI benchmark, GPT-5.6 Sol outperformed its predecessor, GPT-5.5, by scoring 16.2 points higher. This improvement highlights the model’s enhanced ability to optimize and evolve its own architecture without extensive external guidance.
For Japanese markets, where AI-driven technologies are increasingly integrated into financial analysis and trading strategies, advancements like Sol’s autonomous fine-tuning could accelerate automation and innovation across FX, crypto, and equities sectors.