Loading use case index…
Loading use case index…
AI use case
Sakana AI: AB-MCTS - Inference-Time Scaling with Collective Intelligence Sakana AI has developed AB-MCTS (Adaptive Branching Monte Carlo Tree Search), an inference-time scaling …
Core facts from this catalog record. Primary narrative lives in the hero above; full raw fields follow in the next section.
Every column from the source row, in stable order. URLs open in a new tab.
Title
Evolutionary and Collective Intelligence for AI System Development
Content
Sakana AI: AB-MCTS - Inference-Time Scaling with Collective Intelligence Sakana AI has developed AB-MCTS (Adaptive Branching Monte Carlo Tree Search), an inference-time scaling algorithm that enables multiple frontier AI models to cooperate. The research demonstrates that combining different AI models can outperform any single model on complex reasoning tasks. Technical Approach: AB-MCTS extends Monte Carlo Tree Search (famous from AlphaGo) to enable AI to perform effective trial-and-error during inference. The algorithm allows multiple frontier models (like o4-mini, Gemini-2.5-Pro, and DeepSeek-R1) to collaborate by dynamically deciding whether to explore new solutions or refine existing ones. Key Results: AB-MCTS combination of o4-mini + Gemini-2.5-Pro + R1-0528 achieves significantly higher performance on the ARC-AGI-2 benchmark than individual models. This demonstrates that collective intelligence from multiple AI systems can exceed the capabilities of any single system. Significance: The research represents a new direction for AI capability improvements, focusing on inference-time computation rather than only training-time scaling. This approach could enable continued AI capability improvements even as model training costs grow.
Continue exploring AI deployments in the catalog.
Back to use casesCity
Tokyo
Company/Organization
Sakana AI
Continent
Asia
Country
Japan
Category
Research Institution
Type
Deployment
Id
5f06e9b2-e4bf-4a65-ae51-7a4d288de43a
Created At
2026-04-03T18:36:09.905161+00:00