The Test-Time Compute Paradigm: Why Reasoning Models (o1/o3, Claude 3.7 Extended Thinking) Disrupted the Pre-Training Scaling Law

Dr. Julian Vance & Sapiotic Engineering Group

September 5, 2026

Executive Briefing: Shifting From Pre-Training to Deliberative Inference

For seven years, AI capability gains were governed by Kaplan and Chinchilla pre-training scaling laws: spending tens of millions of dollars feeding petabytes of tokens into larger parameter networks. However, with web text corpora largely exhausted, pre-training yield curves began showing diminishing returns. The test-time compute revolution—spearheaded by OpenAI o1/o3 and Claude 3.7 Sonnet Extended Thinking—introduced an entirely new scaling dimension: allocating computational resources during inference time, enabling models to explore, backtrack, and verify chain-of-thought hypothesis trees before generating a final answer.

1. System 1 vs. System 2 Machine Cognition

Daniel Kahneman’s dual-process cognitive framework provides the conceptual foundation for test-time scaling:

Cognitive Dimension Pre-Trained Autoregressive LLM (System 1) Reasoning Test-Time Compute (System 2)
Generation Mechanism Next-token probabilistic guessing; zero error correction. Monte Carlo Tree Search (MCTS), self-critique, backtracking.
Latency & Cost Trade-off Fast (200ms), fixed token cost. Deliberative (5s – 120s), scales with problem difficulty.
Mathematical / Coding Prowess Prone to subtle syntax hallucinations and arithmetic slips. Competitive programming Olympiad gold-medal benchmarks.

2. Process Reward Models (PRMs) & Outcome Reinforcement Learning

Standard Reinforcement Learning from Human Feedback (RLHF) rewards only the final output (Outcome Reward Model). If an answer is correct by pure luck despite flawed reasoning, the model reinforces invalid logic. Process Reward Models (PRMs) evaluate and score every individual intermediate thought step. If step 4 of an equation contains an algebraic hallucination, the search tree immediately prunes that branch, backtracking to explore viable alternative hypotheses.

3. References

  1. Lightman, H., et al. (2023). Let’s Verify Step by Step: Process Supervision in Mathematical Reasoning. OpenAI Research.
  2. Snell, C., et al. (2024). Scaling LLM Test-Time Compute Optimally Can be More Effective than Scaling Model Parameters. UC Berkeley & Google DeepMind.

Leave a Comment