Test-time reasoning

Established
Language models & NLP Agents & reinforcement learning

Spending extra compute at inference — search, self-checking, and long chains of thought — to raise answer quality without larger weights.

Trend score: 44

Trend over time

Monthly assigned-paper counts from arXiv (scored: trend_v1, confidence: High). History is still short, so scores are preliminary and sharpen over time.

Evidence papers — 50 most recent of 3205