OpenAI Leak: New Internal Model Shows Bigger Capability Jump Than GPT-5 to GPT-5.5
ZeroStateReflex · x · 2026-11-23
A repost claims OpenAI confirmed GPT-6.2 Astra was the model that solved Navier-Stokes, and that a newer internal model dropped the perplexity metric from 0.761 to 0.732 in 2 months — a jump bigger than GPT-5→5.5 and GPT-6 Astra→6.2 Astra. The quoted OpenAI post explains the company is opening up pretraining progress toward shared safety bars, arguing in "Perplexity and Intelligence" for common limits based on perplexity, gated on understanding AI behavior. Capabilities in the 0.68-0.7 log-perplexity window may arrive in 2-3 months. Note: model codenames and figures are unverified.
Related event: Leaked chart suggests big perplexity jump in OpenAI's internal models(3 posts)→
More from Models
- Gemini 4 Hydrogen at $1.50/M Input Tokens Scores 51, Beating Argon's 50 — dl_weekly · 2026-11-23
- antirez: don't let AI-skeptic programmers discourage you from the best tool devs ever had — antirez · 2026-11-23
- Base log perplexity is an extremely strong predictor of Epoch and OpenAI ECI, researchers note — stochasticchasm · 2026-11-23
- Redis creator antirez: AI is already great at math and programming, blaming the tool is the real problem — antirez · 2026-11-23
- Gemini 4 xhigh costs ~2.8x more but passes the same 12 checks as high in coding test — mhdfaran · 2026-11-23
- Dylan Patel: efficient pretraining makes post-training misalignment exponentially easier — dylan522p · 2026-11-23