OpenAI reportedly running many pre-trains; far-future model codenamed 'Bel'
haider1 · x · 2026-08-28
The author notes OpenAI has been running a surprising number of pre-trains since late 2025, and discusses the rumor that a big model codenamed Bel is aimed at models much further out.
He speculates OpenAI would tease some of Bel's capabilities along the way, as it did with Astra — especially if they believe RL on Bel could cross their AGI threshold. Note: unconfirmed rumor and speculation.
More from Models
- Dev runs 125B Qwen model locally on M3 Max at 70 tok/s via MLX — mayfer · 2026-08-28
- Qwen3.8-Flash Released: Outperforms DeepSeek and GLM Flash — bindureddy · 2026-08-28
- Thomson-1.0-Small: A RAG-Optimized Fine-tune of Qwen — uber-linny · 2026-08-28
- LLM Logic Test: Minor Value Change Causes Silent Output Failure in OpenAI Models — rayanpal_ · 2026-08-28
- Integrity Bench: A New Benchmark to Measure Model Overconfidence — Acne_Discord · 2026-08-28
- ChatGPT fails to generate anatomically correct human motion diagrams — kaljakin · 2026-08-28