GPT-6.5 'Bel' may be a bigger jump than expected, hinges on pretraining rumors
VraserX · x · 2026-11-01
VraserX speculates that GPT-6.5, codenamed Bel, could deliver a much larger capability jump than people expect. If rumors about its massive pretraining run are true, OpenAI would have a completely different foundation to build on. Pure speculation — no official confirmation yet.
More from Models
- New versions are new models: don't judge GPT Terra and Claude Sonnet by old releases — intellectronica · 2026-11-01
- Forget old impressions: new GPT Terra and Claude Sonnet are excellent — intellectronica · 2026-11-01
- OpenAI now gives ChatGPT subscribers free API credits, but Codex is excluded — Charming_Slide7549 · 2026-11-01
- Memdoor claims 5-model Qwen team beats flagship LLMs on LiveCodeBench Hard — PferdOne · 2026-11-01
- Claude Pro user burns 40% of weekly quota in one day, says $500 tier now needed — lxfater · 2026-11-01
- Why does OpenAI own the entire price-performance Pareto frontier? — mrvoidance · 2026-11-01