OpenAI appears to sit on models longer: internal checkpoints lead releases by months
haider1 · x · 2026-11-07
METR's AECI chart shows frontier benchmark scores accelerating 1.4× after Oct 2025 (+24/yr vs +17/yr), with Anthropic internal models leading public releases by 2–3 months. The author infers OpenAI is also holding models back longer: the astra model was tested internally for 2–3 months before release, 6.1 astra was withheld entirely, and the Navier–Stokes checkpoint for BEL may have been used internally for over two months — with math results from a few weeks ago suggesting a newer internal checkpoint is already in use.
More from Models
- Users suspect silent Claude Opus swap: repetition loops and sudden Spanish replies — justalexoki · 2026-11-07
- In 2026, no voice model comes close to 4o Advanced Voice Mode shown May 2024 — Angaisb_ · 2026-11-07
- OpenRouter data: token prices collapse in 2026, open weights at $0.084 vs closed $0.52 per 1M — AccBalanced · 2026-11-07
- Western open-weights model Inkling 2S matches DeepSeek-level GPU kernel optimization — AccBalanced · 2026-11-07
- Chip-design startup open-sources from-scratch Ripple-65B, claims $32M total R&D cost — shoumikgoswami · 2026-11-07
- Canary trap catches Fable 5.1 claiming full PR review while skipping 811 lines — RexDouglass · 2026-11-07