x.com/scaling01/status/2075269113488789984

OpenAI’s official documentation describes GPT-5.6 Sol and GPT-5.6 Luna as separate members of the same GPT-5.6 family, not as one model post-training the other.

• The GPT-5.6 system card says: “GPT-5.6 is a new family of three models: Sol, our new flagship model; Terra, a capable lower-cost option; and Luna, our fastest and most cost-efficient model.”
• The Sol model page says “GPT-5.6 Sol is the frontier model in the GPT-5.6 family.”
• The Luna model page says “GPT-5.6 Luna is designed for cost-sensitive, high-volume workloads” and that it corresponds to the earlier nano tier.

The likely source of confusion is OpenAI’s PostTrainBench Lite evaluation in the GPT-5.6 system card. But that section does not say Sol post-trained Luna. It explicitly says the benchmark tests a model’s ability to improve “an open-source base model” and lists the actual base models used: Qwen3-4B-Base, Qwen3-1.7B-Base, and SmolLM3-3B-Base. Those are open-source benchmark tasks, not GPT-5.6 Luna.

So the post incorrectly attributes a claim to OpenAI. OpenAI said Sol can improve open-source base models in a benchmark setting; OpenAI did not say GPT-5.6 Sol autonomously post-trained GPT-5.6 Luna.