InSTA: Towards Internet-Scale Training For Agents
This addresses the problem of scalable training for web navigation agents, offering a more efficient alternative to human data collection, though it is incremental in automating data curation with existing LLMs.
The paper tackles the inefficiency of human demonstrations for training web navigation agents by developing a pipeline that uses LLMs to annotate 150k websites with tasks, generate trajectories, and filter them, achieving 97% accuracy in harmful content detection and 82.6% in trajectory success judgment. The resulting agents, based on Qwen 3 1.7B, achieve a 56.9% success rate, outperforming larger models like Qwen 3 235B and reaching 94.7% of Gemini 2.5 Flash's performance.
The predominant approach for training web navigation agents is to gather human demonstrations for a set of popular websites and hand-written tasks, but it is becoming clear that human data is an inefficient resource. We develop a pipeline to facilitate internet-scale training for agents without laborious human annotations. In the first stage, an LLM annotates 150k sites with agentic tasks. In the next stage, LLM agents complete tasks and produce trajectories. In the final stage, an LLM filters trajectories by judging their success. Language models are powerful data curation tools, identifying harmful content with an accuracy of 97%, judging successful trajectories with an accuracy of 82.6%, and producing effective data. We train agents based on Qwen 3 1.7B that are competitive with frontier LLMs as web agents, while being smaller and faster. Our top agent reaches a success rate of 56.9%, outperforming the data collection policy Qwen 3 235B, a 235 times larger Llama 4 Maverick, and reaching 94.7% of the performance of Gemini 2.5 Flash. We are releasing code, models and data at: https://data-for-agents.github.io.