@percyliang: Given how many people have tried their hand at NanoGPT, I really didn't expect a 40% reduction in a single go!
Summary
A new performance record for NanoGPT is announced, achieving a 40% reduction in processing time to 39.9 seconds through optimization at the individual flop level.
View Cached Full Text
Cached at: 09/29/26, 09:52 AM
Given how many people have tried their hand at NanoGPT, I really didn’t expect a 40% reduction in a single go!
Larry Dial (@classiclarryd): New historic NanoGPT record at 39.9s (-27.7s) from @DevenPzak , obliterating the prior record of 67.6s!
This record introduces a new paradigm of thinking to NanoGPT: instead of optimizing matmuls or adding more expressive operations, optimize at the individual flop level with
Similar Articles
NanoGPT Speedrun Frontier
The article reports on autonomous runs comparing 18 frontier AI models on the nanoGPT optimizer speedrun, detailing their performance in closing the gap to the human record.
@eliebakouch: we let opus 4.7 and gpt 5.5 run on the nanogpt optimizer speedrun: ~10k runs, 14k H200 hours, 23.9B tokens. opus hits 2…
AI agents (Opus 4.7 and GPT 5.5/Codex) autonomously optimized the nanoGPT speedrun optimizer, beating the human baseline with a new record of 2930 steps. The blog details their search methods, failures, and releases all run data and code.
Introducing GPT-5.4 mini and nano
OpenAI releases GPT-5.4 mini and nano, smaller, faster variants of GPT-5.4 designed for high-volume workloads with significant improvements in coding, reasoning, and multimodal understanding while maintaining 2x+ faster performance.
@RuiTheBaker: GPT 5.5-level ranking but 27x faster?! @mixedbreadai
Mixedbread's reranker achieves GPT 5.5-level performance on OBLIQ-bench while being 27x faster, according to early results.
Show HN: TurboGPT: train 22KiB transformer in 13s
TurboGPT is an open-source CUDA C++ tool that enables rapid training of a tiny byte-level GPT model, achieving 2.5295 BPB on 1.5G tokens in 13 seconds.