SUMMARYOpenAI is preparing a GPT-6 preview and release cycle in July or August and says frontier AI systems are already automating parts of model training. It also asserts that 10T-class models are being trained by OpenAI, Anthropic, Meta, and xAI, and that AI is solving longstanding problems in software engineering, cybersecurity, and mathematics.
GPT-6 preview and release cycle going through in July August (the new Orion from OpenAI that I was talking about all these prior weeks is apparently GPT-6) ✅
Humanity decisively defeated in the most elite competitive programming competition providing an extremely bullish sign for recursive self improvement ✅
10T class and beyond models being trained by OpenAI, Anthropic, Meta and xAI ✅
OpenAI models had started automating parts of their own model training during GPT-5.3 days in early 2026...now GPT-5.6 SOL is autonomously post-training GPT-5.3 Luna ✅
Countless open problems in SWE, cybersecurity, mathematics etc etc are still getting solved, including 50 year old problems✅
Last year's only and the last human to ever win in a competitive competitive programming competition against frontier AI of the time believes RSI is close ✅
Many OpenAI and Anthropic folks are talking about fully believe that Recursive Self Improvement and a country of geniuses in a data center is right on the horizon...right in front of our eyes ✅
OpenAI claims on livestream that the autonomous AI researcher that we have all been talking about right now is very, very close ✅