Concepttest-time compute scaling
Test-Time Compute Scaling
A method for improving accuracy with fixed model parameters by increasing computational investment during inference, including multiple sampling with voting, extended chains of thought, and search-based decoding.
Core Facts
Timeline (last 90 days)
Sep 4
Investing more computation at the inference stage (test-time compute scaling) can significantly improve model performance, and is often more cost-effective than an equivalent investment of computation at the training stage.
Verified75%