x.com/Algorithon/status/2107287650881208694
1 correction found
Microsoft confirms on publicly accessible web page that OpenAI has been using Looped Transformers in the GPT-6 series
The public Microsoft page for GPT-6.1 Sol does not actually confirm a "looped transformer" architecture. It gives a product description, while OpenAI’s own public GPT-6.1 Sol documentation also avoids making that architecture claim.
Full reasoning
The post turns a speculative architectural inference into a confirmed fact.
What the cited Microsoft page does say publicly is much more limited: Microsoft’s GPT-6.1 Sol catalog entry describes the model as an "efficient frontier reasoning model" in the GPT-6 Sol family and says it has an "efficient serving profile." It does not say "looped transformer," and it does not publicly identify GPT-6.1 Sol’s internal architecture.
OpenAI’s own public GPT-6.1 Sol safety addendum is similarly high-level. It says GPT-6.1 Sol "uses the same types of data and training as GPT-6 Astra" and discusses safeguards/evaluations, but it does not publicly confirm a looped-transformer design either.
So even if someone wants to infer architectural details from other clues, the statement that Microsoft has confirmed OpenAI is using "Looped Transformers in the GPT-6 series" overstates what the public documentation actually establishes.
2 sources
- gpt-6.1-sol | Model Catalog | Microsoft Foundry
GPT-6.1 Sol is OpenAI's efficient frontier reasoning model, designed for complex professional and enterprise workloads... GPT-6.1-Sol is an efficient reasoning model in the GPT-6 Sol family... It delivers strong reasoning capabilities with an efficient serving profile.
- Addendum to GPT-6 Astra System Card: GPT-6.1 Sol - OpenAI Deployment Safety Hub
GPT-6.1 Sol uses the same types of data and training as GPT-6 Astra, described in the GPT-6 Astra card.