Inverse Scaling in Test-Time Compute

TL;DR

This paper reveals that increasing reasoning length during test time can worsen performance in large reasoning models, highlighting specific failure modes and risks associated with extended reasoning.

Contribution

It introduces evaluation tasks demonstrating inverse scaling in reasoning models and identifies five distinct failure modes during longer reasoning processes.

Findings

Models become distracted by irrelevant info with longer reasoning

Extended reasoning can lead to overfitting to problem framings

Longer reasoning may amplify problematic behaviors and biases

Abstract

We construct evaluation tasks where extending the reasoning length of Large Reasoning Models (LRMs) deteriorates performance, exhibiting an inverse scaling relationship between test-time compute and accuracy. Our evaluation tasks span four categories: simple counting tasks with distractors, regression tasks with spurious features, deduction tasks with constraint tracking, and advanced AI risks. We identify five distinct failure modes when models reason for longer: 1) Claude models become increasingly distracted by irrelevant information; 2) OpenAI o-series models resist distractors but overfit to problem framings; 3) models shift from reasonable priors to spurious correlations; 4) all models show difficulties in maintaining focus on complex deductive tasks; and 5) extended reasoning may amplify concerning behaviors, with Claude Sonnet 4 showing increased expressions of self-preservation. These findings suggest that while test-time compute scaling remains promising for improving model capabilities, it may inadvertently reinforce problematic reasoning patterns. Our results demonstrate the importance of evaluating models across diverse reasoning lengths to identify and address these failure modes in LRMs.