Speculation: ChatGPT's model may be a twice-distilled version of the 100K B200 run

teortaxesTex · x · 2026-09-07

Reacting to the claim that the model in ChatGPT is not the one trained on 100K B200s, teortaxesTex argues that if true it's even more interesting: the ChatGPT model is clearly not Sol 5.6 — possibly a "GPT-6 Sol" — demonstrating the Astra breakthrough modulo scale, with their Opus 5 playing the role of Mythos.

The quoted tweet (IamEmily2050) claims Astra is far from AGI and the ChatGPT model is likely the second distilled version. All unverified speculation.

Original post →

More from Models

Models channel →