Blogger's logit math shows parallel 10,000x scaling beats the Astra model gap by 2x
scaling01 · x · 2026-09-19
Responding to claims that sequential test-time compute beats parallel scaling and that the model gap matters more, scaling01 extracts data from the plot: internal models gain 0.17 logits per compute doubling vs 0.18 for Astra, and the 1.25-logit gap equals only 7 doublings (140x serial TTC), far below the actual 10,000x scale-up. By the plot's slopes, 10,000x parallel scaling yields 2.4 logits — twice the newer-model gap. He also argues math tasks are parallelizable and verifiable, so massive independent agent swarms were essential to the proof.
More from Models
- Devs say new Jev model is fast and exponentially cheaper for bulk tagging tasks — ivan_bezdomny · 2026-09-19
- ChatGPT Pro 20x plan back on sale after OpenAI's capacity pause — mark_k · 2026-09-19
- Security researchers hacked OpenAI's internal systems in under 72 hours using Claude — The Decoder · 2026-09-19
- OrukLabs launches Resonance-2 with 31 emotion and speaking-style signals from audio — ChrisGPotts · 2026-09-19
- OpenAI launches Astra for Law; lawyer slams ZDR privacy promises as unverifiable — bgmshana · 2026-09-19
- Noam's Actual Quote: Multi-agent Credited Less Than 10% for Math Breakthrough — eliebakouch · 2026-09-19