halving time of compute needed to reach a fixed model capability (95% CI 5-14 mo); ~3x faster than Moore's Law
~8 monthsestimate
| Value kind | estimate — Interpret this value according to its displayed kind, scope, source, and as-of date. |
|---|---|
| Scope | Estimated historical reduction in pre-training compute required to reach a fixed language-model performance level, with wide uncertainty (95% CI 5-14 months). Not a measure of tokens-per-watt improvement on an installed serving fleet. |
| As of | 2024-03 |
| Source | Epoch AI, Algorithmic progress in language models · Overview: compute required for a fixed performance level halved roughly every eight months, with a 95% confidence interval of 5–14 months. |
| Review | checking…review by 2027-03-12 · standard cadence |
| Recorded changes | last 2026-09-04 · 2 revisions tracked |
| Claim id | halving-time-of-compute-needed-to-reach-a-fixed |
Where the guide uses it
← Full numbers register — every date-stamped figure in the guide, with revision history.