MidTool: Mid-training Data Synthesis for Agentic Tool Use

Source

• Source name: arXiv
• Source kind: paper
• Authors / organization: Fengqing Jiang, Yite Wang, Boyi Liu, Zhaoyang Wang, Canwen Xu, Zhewei Yao, Radha Poovendran, Yuxiong He
• Venue: arXiv: cs.AI
• Published: 2026-08-20
• Collected: 2026-08-21
• URL: http://arxiv.org/abs/2608.20314v1
• PDF: https://arxiv.org/pdf/2608.20314v1
• DOI: UNKNOWN
• Registry group: academic_apis_and_preprints
• Evidence role: canonical_academic_index
• Language: en
• Country / region: UNKNOWN

Plain-Language Brief

In the paper 'MidTool: Mid-training Data Synthesis for Agentic Tool Use', researchers present a new open corpus construction pipeline called MidTool. This pipeline is designed to teach large language models how to use tools effectively by recognizing tool affordances, grounding arguments from context, composing tool call workflows, and recovering from incomplete information. The authors mid-trained Qwen3-4B-Base and Qwen3-8B-Base on MidTool-Mix, a combination of various data sources, and then fine-tuned and reinforced the models. The results indicate that models trained with MidTool perform better than baselines in downstream tasks related to tool use.

What the Document Says

• MidTool is a pipeline for mid-training large language models in general tool use.
• The pipeline combines various data sources: web, PDF, and code data, as well as synthesized supervision from real-world tool APIs, MCP skills, and document-grounded workflows.
• MidTool aims to teach models how to recognize tool affordances, ground arguments from context, compose tool call workflow, and recover from incomplete information.
• The authors mid-trained Qwen3-4B-Base and Qwen3-8B-Base on MidTool-Mix.
• Post-training was done using both supervised fine-tuning and reinforcement learning.
• Comparisons with baselines show that MidTool-trained models consistently perform better in downstream tasks related to BFCL, tau2-Bench, and MCP Universe.

Key Concepts Mentioned

• MidTool
• mid-training
• large language models
• general tool use
• tool affordances
• context
• workflow
• Qwen3
• downstream tasks

Claims Made by the Document

• MidTool improves downstream performance in various tasks.
• Dedicated mid-training benefits general tool use.

Evidence Used

• Results from comparisons with baselines.

Methods / Approach

• Mid-training with MidTool pipeline, followed by post-training with supervised fine-tuning and reinforcement learning.

Limitations Stated by the Authors

Not stated in source.

Limitations Noted by the Agent

Agent interpretation — not from the source.

The results may be specific to the Qwen3 models and the datasets used (BFCL, tau2-Bench, MCP Universe).

Open Questions

• How does MidTool compare with other methods for improving general tool use in large language models?
• Can MidTool be adapted to improve tool use in other types of AI models?

Notes for Future Q&A

May include interpretation. Labeled where it does.

The paper focuses on the benefits and methodology of using MidTool for mid-training large language models in general tool use. The results suggest that this approach improves performance in downstream tasks, but further research is needed to compare it with other methods and evaluate its applicability to other types of AI models.

Raw Abstract / Excerpt

Mid-training is increasingly recognized as a critical stage for shaping the capabilities of large language models. Recent work has shown that targeted mid-training can strengthen reasoning-intensive abilities such as math and science, and can also improve agentic capabilities in software-engineering settings. In this work, we study the parallel but less explored agentic capability: general tool use. We present MidTool, an open corpus construction pipeline for agentic tool-use mid-training that combines large-scale web, PDF, and code data with synthesized supervision from real-world tool APIs, MCP skills, and document-grounded workflows. MidTool is designed to teach models how to recognize tool affordances, ground arguments from context, compose tool call workflow, and recover from incomplete information. We mid-train Qwen3-4B-Base and Qwen3-8B-Base on MidTool-Mix, and then apply follow-up post-training with both supervised fine-tuning and reinforcement learning. Compared with baselines, MidTool-Mix consistently improves downstream performance under both SFT and RL on BFCL, tau2-Bench, and MCP Universe. These results suggest that general tool use, like other important LLM capabilities, benefits from dedicated mid-training rather than being left entirely to post-training.