Some Codex users suspect GPT‑6 Sol is a renamed older model
A post arguing that Sol feels like an upgraded Terra drew 24,000 views, though a leaderboard run puts it 72 points above its own predecessor.
An account posting as notjazii said on Wednesday that GPT-6 Sol does not feel like Sol at all, and that it feels closer to Terra. The account said it had used the model since release and that the price was almost the same as GPT-5.6 Terra. The post had 322 likes and 24,565 views.
The argument rests on an inference rather than evidence. If Sol was distilled from GPT-6 Astra, the account asked, should it not sit closer to Astra. It said it might be completely wrong, and suggested a real Sol upgrade could arrive at OpenAI's developer day next Tuesday.
A similar joke had circulated on Tuesday. Another account wrote a mock pitch in which OpenAI takes GPT-5.6 Sol and Luna, cuts the price by half, ships them under a new name, and leaves subscription token allowances untouched.
One public measurement cuts against the suspicion. Code Arena said GPT-6 Sol at maximum effort took fourth place in its WebDev ranking with 1,689 points, an improvement of 72 points on GPT-5.6 Sol, which had ranked seventeenth. It said the model moved from eighteenth to third on simulations.
OpenAI has not addressed the claim and has published no architecture detail for Sol. Code Arena's ranking measures output rather than what produced it, so a score can rise without settling what the model underneath is. The suspicion is a user's, and it remains unconfirmed.