AI Paper Atlas

Research map / Multi-agent systems and games / LLM agents

MACoT: Synthesizing Chains of Thought for Small Models via Multi-Agent Collaboration

Guokai Tang, Feng Zhao

AAAI 2026 · Multi-agent systems and games › LLM agents · approach: Agent data synthesis and behavior studies · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Proposes MACoT, a multi-agent collaboration framework that synthesizes chains of thought for small language models as an alternative to distilling long CoTs from large reasoning models.

Problem

Distilling long chains of thought from large reasoning models into small language models is common but the abstract cuts off before stating the specific shortcomings; only the motivation (SLMs are efficient but need better reasoning) is stated.

Approach

Synthesizes chain-of-thought training data for small models through collaboration among multiple agents rather than copying long-CoTs from large reasoning models. Details beyond the abstract are not available.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add the Problem Passport analysis stage (ablation)MACoT · AIMEPass@1 accuracy (%): 8.3 -> 11.67Improved
add the Problem Passport analysis stage (ablation)MACoT · AMC23Pass@1 accuracy (%): 45.0 -> 52.5Improved
add one self-reflection paragraph via Student-Teacher refinement (ablation)MACoT · BBHPass@1 accuracy (%): 61.5 -> 63.64Improved
add one self-reflection paragraph via Student-Teacher refinement (ablation)MACoT · MATH500Pass@1 accuracy (%): 76.6 -> 77.4Improved
increase self-reflection paragraphs beyond one (N2, N4) (ablation)MACoT · AIME24 and MATH-500Pass@1 accuracy (%): N2 average drop 1.8 points, N4 average drop 2.4 pointsHurt
synthesize CoTs with MACoT instead of using original LIMO CoTs (same 817 problems)Qwen2.5-7B-Instruct · BBHPass@1 accuracy (%): +2.68 pointsImproved
synthesize CoTs with MACoT instead of using original LIMO CoTs (same 817 problems)Qwen2.5-7B-Instruct · GPQA and Omni-MathPass@1 accuracy (%)Hurt
fine-tune on MACoT CoTs instead of R1 long-CoTs (same 1879 problems)Qwen2.5-7B-Instruct · Average of 9 benchmarksAvg. accuracy (%): 42.2 -> 50.04Improved
fine-tune on MACoT CoTs instead of R1 long-CoTsQwen2.5-7B-Instruct · Average of 9 benchmarksAvg. tokens: 4972 -> 745Improved
fine-tune on MACoT versus base modelQwen2.5-7B-Instruct · GaokaoPass@1 accuracy (%): 41.31 -> 41.31No clear effect
fine-tune on Numina CoTs versus base modelQwen2.5-7B-Instruct · Average of 9 benchmarksAvg. accuracy (%): 46.8 -> 39.7Hurt
fine-tune on R1 long-CoT distillation versus base modelQwen2.5-7B-Instruct · Average of 9 benchmarksAvg. accuracy (%): 46.8 -> 42.2Hurt

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy