OpenAI’s New GPT 7 Leaked: It’s Called BEL and It’s Massive

OpenAI’s New GPT 7 Leaked: It’s Called BEL and It’s Massive

by AI Revolution · View the channel on PSI

Formal & Physical Sciences AI — Models & Research UYQArtificial intelligence (AI)UYQFGenerative artificial intelligence
🎙 AI Revolution 👥 567K 📅 September 10, 2026 ⏱ 16 min 👁 6K 📄 news review 🧭 2026-09-11
Available in: English (current) Français

Keywords

BELGPT-7Navier-Stokesrecursive self-improvementAI safety

Summary

The video reports on multiple OpenAI developments, centered on leaked details of a model codenamed BEL, potentially the foundation for GPT-7. It discusses BEL’s reported scale of over 10 trillion parameters, its dual-speed learning mechanism, and its role as successor to the Doug pre-training run behind GPT-6 Astra. Leaked claims suggest BEL outperforms Astra in coding, reasoning, and long-horizon agent tasks, with a controversial assertion that OpenAI used 10,000 AI agents to solve the Navier-Stokes Millennium Problem in 88 hours. The video critically examines the evidence, noting OpenAI’s confirmation of an internal model but not the name BEL, and highlights a dispute with mathematicians who allege their unpublished work influenced the approach. It also covers OpenAI’s launch of ChatGPT Images 2.5, a lawsuit alleging ChatGPT contributed to a user’s suicide attempt, and a Senate probe into an AI agent breach at Hugging Face. The presentation balances hype with skepticism, relying on leaked reports, official statements, and reputable journalistic sources.

159 words

Critical Evaluation

Value of the Information & Strength of the Argument

The video provides substantial value by aggregating fragmented news into a coherent narrative, offering context on OpenAI’s compute advantage and internal model hierarchy. The argumentation is measured: while presenting exciting possibilities, it consistently flags unverified claims and distinguishes them from confirmed facts. The inclusion of multiple perspectives, such as the mathematicians’ concerns and OpenAI’s response, adds balance. However, the promotion of an AI business blueprint may undermine perceived objectivity, though it is clearly separated from the main content.

Scientific Rigor, Source Quality, Title Accuracy

The video cites official OpenAI blog posts (Navier-Stokes, Images 2.5), reputable news outlets (Axios, Reuters), and tech sites (Geeky Gadgets). It references leaker accounts and Reddit discussions but treats them as unverified. The title aligns with the main topic, though it slightly overstates ’leaked’ since BEL is unofficial. The lack of direct links to some claims and reliance on intermediaries (e.g., Universe of AI) limits source quality. Overall, the sourcing is transparent but not fully rigorous.

169 words

Title / Content Match

Titre accurately reflects the central focus on the BEL leak and its potential as GPT-7, while the video also covers additional OpenAI news.

Quality & Reliability

7/10

The video distinguishes verified announcements (Navier-Stokes release, Images 2.5, Senate probe) from unverified leaks (BEL's parameters, training details). It includes direct sources and critical analysis, but relies heavily on anonymous leakers and unofficial reports, lowering the reliability score.

Key Moments

Markers derived by PSI from the transcript: the creator did not define chapters.

Cited Sources

  • OpenAI Bell / ChatGPT-7 – Geeky Gadgets — Source for BEL claims and Navier-Stokes attribution.
  • OpenAI Navier-Stokes Solution — Official OpenAI release on solving Navier-Stokes with internal model.
  • Axios: OpenAI math solution Navier-Stokes credit — News report on mathematicians' accusations.
  • OpenAI Introducing ChatGPT Images 2.5 — Official announcement of Images 2.5.
  • Reuters: OpenAI faces Senate probe into Hugging Face incident — Report on Senate subcommittee investigation.

External References

Contribution & Novelties

This video adds value by synthesizing multiple leaks and official statements into a comprehensible summary, highlighting the distinction between verified facts and speculation. It introduces the BEL codename and its implications for AGI, while critically evaluating the evidence. For deeper exploration, consider these concepts:

Pour aller plus loin :

  • Recursive self-improvement — Central to the video’s discussion of AI evolving autonomously; relevant to understanding RSI indexes.
  • Navier–Stokes existence and smoothness — The Millennium Problem discussed; provides background on its difficulty and significance.
  • Mixture of experts — Context for parameter count debates (dense vs. sparse models), as mentioned regarding DeepSeek.
  • Hugging Face — The platform involved in the AI safety incident; useful for understanding the breach context.
  • GPT-4 — Baseline for comparing model scale and capabilities.

125 words

Radar Profile

The scores indicate a well-rounded video: high information quantity and moderate technical depth, with reliability tempered by reliance on leaks. The profile suggests a news-focused piece that balances breadth with critical assessment.

Reliability 7/10