The record for training a GPT-2-class model to the NanoGPT speedrun target fell to 88.1 seconds, announced, like the December and March records, in a post from the same X account (@classiclarryd). The step down from the 99.3-second bigram-hash record five weeks earlier is 11.2 seconds, the largest single drop the corpus has logged on a ladder that began here at 127.7 seconds in December; it is overtaken eight days later by 86.8 seconds. The newsletter files it under training curves still compressing, two days before the AdderBoard minimum fell from 121 to 36 parameters.
Development
The speedrun record falls to 88.1 seconds
The corpus's longest-running ladder, the time to train a GPT-2-class model, takes its largest single step down so far, which the author files under training curves still compressing.
Claim
The NanoGPT speedrun record dropped to 88.1 seconds.
Reported metric: 88.1 s
Reporting date: (not necessarily the event date)
Sources
Attribution and reporting linked by this record; links do not imply independent verification.
- Welcome to February 28, 2026 (theinnermostloop.substack.com)
- Post on X: NanoGPT speedrun record of 88.1 seconds (x.com)
Verification record
These entries record a check, not proof that the underlying claim is true. Unless described in the article below, the scope of the check is unknown.
- Recorded by claude-fable-5-1/2026-09-17 on 2026-09-17T08:00:00Z
Cite or correct this record
Suggest a correction — opens a GitHub draft for you to review and submit.