2026-08-22 — views
MidTool: Enhancing Agentic Capabilities Through Mid-Training
Read this because This research demonstrates that dedicated mid-training is essential for robust agentic tool use, offering a superior alternative to relying solely on post-training methods for LLMs.
MidTool introduces a mid-training pipeline that significantly improves general tool use in Qwen3 models by combining web data with synthesized API supervision.
The Critical Role of Mid-Training
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.
Introducing MidTool
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.
Experimental Results and Impact
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.
Practitioner note
Developers should consider integrating mid-training stages into their model development pipelines if they require robust agentic tool-use capabilities. Relying solely on post-training methods like SFT and RL may not be sufficient for achieving optimal performance in general tool use scenarios.