Alibaba Unveils Marco-o1

A Groundbreaking LLM for Complex Problem-Solving

In partnership with

Alibaba’s MarcoPolo team has introduced Marco-o1, a cutting-edge large language model (LLM) designed to tackle both traditional and open-ended problem-solving tasks with unparalleled accuracy. Building on advancements from OpenAI’s o1 model, Marco-o1 integrates innovative techniques to push the boundaries of AI reasoning, especially in areas like mathematics, physics, coding, and tasks with ambiguous standards.

Sponsored
NeuroTycoonGet easy tips to make your brain better and win in business, in just 3 minutes a week.

A standout feature of Marco-o1 is its incorporation of Chain-of-Thought (CoT) fine-tuning, Monte Carlo Tree Search (MCTS), and a novel reflection mechanism. These technologies work together to enhance the model’s problem-solving capabilities, enabling it to navigate complex reasoning challenges. The model’s training involved over 60,000 carefully curated examples from diverse datasets, including a specialized Marco Instruction Dataset, ensuring robust performance across different domains.

Marco-o1 has shown remarkable progress in multilingual tasks, with a notable 6.17% improvement in English MGSM dataset accuracy and 5.60% for its Chinese counterpart. The model excels in translation, particularly in capturing colloquial expressions and cultural nuances.

One of its most innovative aspects is its use of varying action granularities within the MCTS framework, allowing for both broad and finely detailed reasoning. This flexibility, combined with the model’s reflection mechanism, enables Marco-o1 to re-evaluate and refine its answers, improving accuracy in complex scenarios.

While still a work in progress, Alibaba’s Marco-o1 represents a significant step forward in AI’s ability to solve intricate problems. The model and its datasets are available on GitHub, inviting further research and development within the AI community.

Start learning AI in 2025

Everyone talks about AI, but no one has the time to learn it. So, we found the easiest way to learn AI in as little time as possible: The Rundown AI.

It's a free AI newsletter that keeps you up-to-date on the latest AI news, and teaches you how to apply it in just 5 minutes a day.

Plus, complete the quiz after signing up and they’ll recommend the best AI tools, guides, and courses – tailored to your needs.