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

Sakana AI just dropped a game-changer: an open-source algorithm that turns AI models into a collaborative dream team. Forget solo acts; their Adaptive Branching Monte Carlo Tree Search (AB-MCTS) algorithm lets AIs tackle complex problems together like never before. Think of it as adding a third dimension to AI thinking. When a challenge pops up, this algorithm doesn’t just ponder longer or explore wider; it intelligently handpicks theperfectAI model for the job. And if things get really hairy? It unleashes a whole squad of AI powerhouses, working in sync to crack the code.

Sakana AI Releases Algorithm That Makes AI Models Think Collectively

Imagine a digital hive mind where AI giants like Gemini 2.5 Pro and DeepSeek-R1 pool their intellect. A Tokyo AI lab just flipped the switch on this future with its inference-time scaling algorithm, creating a collaborative playground where the world’s most powerful AI models learn and evolve together.

Sakana AI isn’t just tinkering with artificial intelligence; they’re orchestrating an evolution. Their quest: to forge a superior AI by artfully blending the strengths of individual models while purging their inherent biases. Years of research culminated in a groundbreaking 2024 paper unveiling their method: “evolutionary model merging,” a technique poised to redefine AI performance.

Imagine AI that adapts on the fly. This company just unleashed an algorithm that does exactly that. Think smarter decisions: AI models now juggle processing power, explore diverse solutions by generating multiple outputs, and even team up to tackle complex problems. The result? Peak performance, budget-friendly operation, and a whole new level of AI ingenuity.

Sakana AI’s AB-MCTS system, a chimera of o4-mini, Gemini 2.5 Pro, and R1-0528, didn’t just meet expectations on the ARC-AGI-2 benchmark – it shattered them. Individually, o4-mini could crack 23% of the problems. But when fused within the AB-MCTS framework, that number jumped to 27.5%, proving that sometimes, the sum is greater than its parts in the quest for artificial general intelligence.

Sakana AI just dropped a bombshell: TreeQuest is live on GitHub! But that’s not all – brace yourselves for a glimpse into their ARC-AGI experiments, now unveiled. Dive deep into the research detailed in their newly released arXiv paper. The future of AI is here, and it’s unfolding before our eyes.

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