Sakana AI Releases Open-Source Algorithm That Lets Multiple AI Models Collaborate on Complex Tasks

Dubbed Adaptive Branching Monte Carlo Tree Search (AB-MCTS), it is a new inference-time scaling algorithm by Sakana AI.

Written by Akash Dutta, Edited by Siddharth Suvarna | Updated: 3 July 2025 19:02 IST

Highlights

Sakana AI found the algorithm to boost overall performance on benchmarks
The algorithm works on the principle of test-time scaling
It allows models to perform Sequential Refinement and Repeated Sampling

Sakana AI Releases Open-Source Algorithm That Lets Multiple AI Models Collaborate on Complex Tasks

The open-source algorithm is available to download via Sakana AI’s GitHub listing

Photo Credit: Unsplash/Gerard Siderius

Sakana AI released an open-source algorithm on Tuesday, which allows multiple artificial intelligence (AI) models to collaborate on complex problems. Dubbed Adaptive Branching Monte Carlo Tree Search (AB-MCTS), it is an inference-time scaling or test-time scaling algorithm that adds a third dimension to the existing framework of AI models. With this, when faced with a new problem, the system not only decides if longer reasoning is suitable or wider exploration, but it also decides which AI model is best suited for the task. In case the problem is too complex, it can also deploy multiple AI models.

Sakana AI Releases Algorithm That Makes AI Models Think Collectively

In a post on X (formerly known as Twitter), the Tokyo-based AI firm highlighted that its new inference-time scaling algorithm creates an environment for collective intelligence for AI by letting frontier models such as Gemini 2.5 Pro, o4-mini, and DeepSeek-R1 to collaborate.

The company set out to solve a complex problem in the AI domain — how to combine the unique strengths and eliminate the unique biases of AI models to achieve higher performance. Sakana AI has been researching this problem for multiple years, and in 2024, it published a paper on “evolutionary model merging.”

Now, building on its findings, the company has released an algorithm which creates a system that lets AI models perform test-time compute on specific budgets, lets them generate multiple outputs to explore different perspectives, and even put multiple AI models suitable for the task to achieve higher performance.

Researchers working on the project were also able to test the capability on the ARC-AGI-2 benchmark, where the AB-MCTS system used a combination of o4-mini, Gemini-2.5-Pro, and R1-0528, and was able to surpass the performance of the individual models. Sakana AI claimed that while o4-mini solved 23 percent of the problems independently, it reached 27.5 percent when it was part of the AB-MCTS cluster.

Sakana AI has released the TreeQuest algorithm on its GitHub listing and has also shared its ARC-AGI experiments separately. The details from the study have been published in a paper on arXiv.

For the latest tech news and reviews, follow Gadgets 360 on X, Facebook, WhatsApp, Threads and Google News. For the latest videos on gadgets and tech, subscribe to our YouTube channel. If you want to know everything about top influencers, follow our in-house Who'sThat360 on Instagram and YouTube.