Inverse scaling – Definition & Meaning

Definition of inverse scaling

: a pattern in which performance on a particular AI task becomes worse as model scale increases, such as when larger models in a comparable family perform less accurately than smaller ones
- The benchmark showed inverse scaling on a subset of spatial questions even though the larger model had a higher overall score.
- The researchers tested additional model sizes to see whether the inverse scaling trend continued or eventually reversed.

Origin & history

The Inverse Scaling Prize and the 2023 paper Inverse Scaling: When Bigger Isn't Better helped establish the term in language-model evaluation. Later work examined reversals at larger scales and, in October 2026, task-specific inverse scaling in vision-language models.

Editorial references

Sources & further reading

3 sources

Inverse Scaling: When Bigger Isn't Better Research Paper · arXiv · 2023-06-15
Inverse Scaling Can Become U-Shaped Conference Paper · EMNLP / ACL Anthology · 2023-12
On the Limits of Scale: Inverse Scaling in Vision-Language Models Research Preprint · arXiv · 2026-10-19

Test yourself

Which of these is the meaning of inverse scaling?

Cite this entry

"Inverse scaling." AI Dictionary, Dadgogo, https://dadgogo.com/dictionary/inverse-scaling/. Accessed 21 Oct. 2026.