rasbt debunks Astra looped-transformer hype: fewer tokens likely mean smarter model, not hidden reasoning
rasbt · x · 2026-09-04
Responding to The Information's report that OpenAI's Astra uses a 'recurrent depth or looped transformer' architecture, Sebastian Raschka argues the looped design isn't hiding reasoning tokens. GPT 6 Astra using fewer tokens than GPT 5.6 Sol likely reflects it being a smarter, bigger model. He notes the same pattern within prior generations: GPT 5.6 Sol uses 46% fewer output tokens than GPT 5.6 Luna on the intelligence index, yet nobody claims Sol hides its reasoning more.
Related event: OpenAI's Astra Reportedly Uses Recurrent Depth, Raising AI Safety Concerns(7 posts)→
More from Models
- Fable 5.5 delayed: 'Stoptober' meme says no one dares ship a SOTA model now — patience_cave · 2026-10-21
- Meta details 2026 GEM ads model: Kunlun architecture brings 2.5x complexity growth — Meta_Engineers · 2026-10-21
- Decision Models' routing classifier hits AntSeed, cutting inference cost to 24% of always-Opus — airesearch12 · 2026-10-21
- Developer flags anomalous Claude Projects usage: 98% weekly limit consumed at 1% thread share — 0xkarasy · 2026-10-20
- DeepSeek hands inference providers 90%+ margins, pressuring Moonshot and Zhipu hosting deals — teortaxesTex · 2026-10-20
- HelixWorld world model tested: 70GB VRAM per BF16 GPU run, consumer cards need not apply — cocktailpeanut · 2026-10-20