On Anthropic’s standing test of speeding up model-training code, Mythos Preview reached roughly 52x against the 4x a skilled human engineer reaches and the 3x that Opus 4 managed (the post dates it to 2024; the model shipped in May 2025) (Anthropic on X; report). The same thread reports Anthropic engineers shipping 8x as much code per quarter as in 2021-2025. It is the direct successor of February’s 34x speedup by Opus 4.6, where 4x had counted as a day’s human work, and it sits beside the 400x internal research speedup claimed for Claude Mythos in April; the steering result from the same report covers the problem-selection half of the loop.
Development
A model speeds up training code thirteen times better than a skilled human
The clearest instance in the corpus of a model outperforming skilled humans at the very task that speeds up making the next model, which is the mechanism the recursive loop depends on.
Claim
On a standing test to speed up model-training code, Mythos Preview reached roughly 52x where a skilled human reaches 4x and 2024's Opus 4 managed 3x, while Anthropic engineers now ship eight times as much code per quarter as in 2021-2025.
Reported metric: ~52x vs 4x human
Reporting date: (not necessarily the event date)
Sources
Attribution and reporting linked by this record; links do not imply independent verification.
- Welcome to June 5, 2026 (theinnermostloop.substack.com)
- Anthropic on X: Mythos Preview hits ~52x on the training-code speedup test (x.com)
- Anthropic on X: engineers ship 8x as much code per quarter (x.com)
- When AI builds itself (www.anthropic.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.