We watch frontier labs for silent cuts, so you don't have to. Mid models stay off this page. If a flagship gets quieter, it lands here — with the pelican canary and the indexed writeup.
Lab named three defects and reset usage. House pelican not re-run after the fix.Sep 12
Same post-launch cycle in July. No sourced claim that it is still happening.Jul 20
No sourced incident on the ledger.—
Launch pelican was clean. No nerf claim since.Sep 1
Frontier, watched. Nothing on the book.—
Launch pelican was strong. No later cut on the ledger.Jun 29
Frontier, watched. Nothing on the book.—
confirmed alleged pelican
Same prompt. Frontier only. House test is ours.
Launch-day grid: every Astra level beat the best Sol pelican. Low already looked like a bird on a bike.
strong OpenAI · Sep 12Quality complaints a week later. House re-run not posted yet. Slot reserved.
retest Anthropic · Sep 1Launch-day animated pelican. Clean frame, pouch, pedaling.
strong Google · Jun 29Called the best pelican-on-a-bike SVG Willison had seen from any model at the time.
strongFamily grid of 18 pelicans. Even xhigh Sol stayed abstract next to later Astra drawings.
midFrontier, watched. No house re-run on the board yet.
pendingFrontier, watched. Not on this board until we or a cited source run the exact prompt.
pendingFrontier, watched. Slot reserved for a house run.
pendingGoogle-indexed writeups and sourced posts. Frontier only.
A week after launch, identical-prompt tests look worse. OpenAI later named three defects, including a context experiment on 4,000–5,000 users, and reset Codex usage. Sol went through the same cycle in July.
Decrypt → Yellow →
synthwavedd called it a post-launch lobotomy. Pranjal Paliwal took back an AGI take after reading the code. Salio and Md Ismail Sojal both reran identical prompts. Dax Raad’s Opencode team switched back to Sol.
Decrypt →
Simon Willison’s launch-day SVG canary. Every Astra pelican, from low to max, looked better than the best Sol drawing. Astra low cost 9.55¢ and still won.
simonwillison.net →
Launch-day Fable 5.1 SVG. The bird is a pelican, the bike is a bike, the pouch is a pouch.
simonwillison.net →
Launch fireworks, then a week of “it got dumber” posts. That is the pattern this site exists to timestamp.
Decrypt →
Informal, no score — one of the few public canaries that survives a launch week.
AI Weekly →