Reasoning & Test-Time Compute · 2024

Scaling LLM Test-Time Compute Optimally can be More Effective than Scaling Model Parameters

Charlie Snell, Aviral Kumar

Establishes 'compute-optimal' test-time scaling, allocating inference compute by problem difficulty so a small model can beat a 14x-larger one, framing the test-time-compute axis.

Editorial record

Plain-language summary

Establishes 'compute-optimal' test-time scaling, allocating inference compute by problem difficulty so a small model can beat a 14x-larger one, framing the test-time-compute axis.

Knowledge graph

Relationships

Descendants

Source record

Provenance

Record ID
P-631
Record created
2026-07-13
Last reviewed
2026-07-14
Record version
2

Citation caveat: Citation metadata is approximate and marked unverified in the source dataset.