Test-Time Compute
The allocation of additional processing power and time during inference to allow a model to search, plan, or reason before outputting a final answer.
Think of It Like This
Like giving a student two full hours to carefully outline and draft an essay, rather than forcing them to answer in thirty seconds.
Traditionally, AI models scale by increasing training compute. Test-time compute scales intelligence during deployment. Models like OpenAI's o1 utilize complex internal reinforcement learning and search algorithms (like Monte Carlo Tree Search) during generation. Giving these models more time to 'think' exponentially improves their logical accuracy.