Your API swapped models on you this week. No changelog, no announcement. Multiple users picked GPT-5.6 Sol and got responses stamped gpt-6-sol. Packet captures on Codex show the same swap. A V2EX thread pulled 2,700+ views documenting it. O...
Your API swapped models on you this week. No changelog, no announcement.
Multiple users picked GPT-5.6 Sol and got responses stamped gpt-6-sol. Packet captures on Codex show the same swap. A V2EX thread pulled 2,700+ views documenting it. One tester ran the same prompt on both: 72k tokens in 9 minutes on the new one, 26 minutes on the old.
Yesterday the rumor mill had a whole "GPT-6 Sol Thursday?" cover ready. Today Altman moved his "big launch" to next week instead. Tibo's take: "Sometimes physics can't be cheated."
OpenAI hasn't confirmed the model name. But read together with the $200/mo Pro pause ten days ago, the picture is pretty clear: GPT-6 Sol is likely already serving traffic in grayscale, and the official launch slipped because capacity is tight.
The model itself is half the story. Man, a frontier lab can reroute production traffic to a different model and you find out from a model_id field.
Do you pin model versions in prod, or ride the default? Genuinely curious.