GPT-7 & GPT-8 Leaks: OpenAI Announces AGI Threshold
Even as most of us are still getting used to GPT-6 Astra and GPT-6.1 Sol, GPT-7 and GPT-8 leaks show that development is going on at a rapid pace at OpenAI. First I explain what OpenAI's head of applied research, Boris Power, actually said about 80 to 90 percent of OpenAI's research being aimed at GPT-7, GPT-8 and beyond. Then we look at the Bel leak, a reported 10 trillion parameter base model built for the generation after GPT-6. After that, we go through the machine being built to train these models, from Stargate and the Abilene site to the MRC networking protocol and OpenAI's first custom chip. Next, we discuss how OpenAI's own AI agents now put in 3.1 workdays for every human researcher workday, and why that could speed up GPT-7. Then I explain why safety sets the pace, and what GPT-7 and GPT-8 could look like. Finally, I share my honest take on what is confirmed and what is still a leak. This is an important part of my ongoing quest to find the best AI model in 2026. Subscribe to the channel, as I will break down every new OpenAI leak and the official GPT-7 news the day it lands. Timestamps: 0:00 - GPT-7 and GPT-8 Leaks 0:58 - The 80 to 90 Percent Claim 3:13 - The Bel Leak 4:55 - The Machine Behind GPT-7 7:18 - AI on the Research Team 9:06 - Why Safety Sets the Pace 10:33 - What GPT-7 and GPT-8 Could Be 12:28 - My Honest Take #Astra #GPT7 #GPT8 #OpenAI #ChatGPT #GPT6 #SamAltman #AGI #AINews #ArtificialIntelligence #AIModels #FrontierAI #ClaudeAI
