ReasonForge

Interactive Test-Time Compute Laboratory

live compute engine Game-of-24 corpus · puzzles tokens burnt · 0 solved · 0/0

Scaling Lab — the inference-compute frontier

Sweep the test-time compute budget and watch every strategy trace its own scaling curve on the Game-of-24 benchmark. This is the diagram behind Scaling LLM Test-Time Compute Optimally (Snell et al., 2024) — but live and interactive.

Controls
Strategies on chart
Greedy (1×)
Best-of-N + ORM
Self-consistency (CoT-SC)
Tree-of-Thoughts (BFS)
MCTS + PRM
Self-Refine
What to look for: below ~23 compute all strategies hug greedy; between 24–26 search methods pull ahead; MCTS+PRM maintains the highest slope because process rewards prune bad trajectories cheaper than outcome rewards. This is the core empirical finding of 2024–2026 inference scaling work.