TLDR: Microsoft has announced the public preview of Deep Research within its Azure AI Foundry Agent Service. This new offering leverages advanced GPT-powered agents and Bing Search to automate complex, multi-step web research, analysis, and synthesis for enterprises, providing transparent and auditable outputs.
Microsoft has officially unveiled the public preview of Deep Research, a significant new capability integrated into its Azure AI Foundry Agent Service. Announced around July 8-15, 2025, this offering is set to transform how enterprises conduct complex web research by enabling advanced automation through composable, GPT-powered agents.
Deep Research is designed to move beyond traditional information retrieval, allowing AI agents to autonomously plan, analyze, synthesize, and report on vast amounts of web-based information. It delivers OpenAI’s advanced agentic research capabilities as an API and SDK solution, seamlessly integrated with Azure’s enterprise-grade agentic platform. This empowers developers to embed sophisticated research automation directly into applications, workflows, and multi-agent systems.
According to Yina Arenas, VP of Product, Core AI, at Microsoft, ‘With Deep Research, developers can build agents that deeply plan, analyze, and synthesize information from across the web—automate complex research tasks, generate transparent, auditable outputs, and seamlessly compose multi-step workflows with other tools and agents in Azure AI Foundry.’
The core of Deep Research operates through a multi-stage agent pipeline, meticulously designed to mimic rigorous human research workflows. This pipeline includes:
Intent Clarification: Utilizing advanced GPT-series models, such as GPT-4o and GPT-4.1, to precisely scope the initial research query.
Web Data Discovery: Securely invoking the Grounding with Bing Search tool to gather high-quality, recent web data, which significantly mitigates hallucination risks and ensures factual accuracy.
Report Generation and Traceability: Culminating in a structured, source-cited report that not only provides the final answer but also meticulously documents the model’s step-by-step analytical path, including all utilized citations and any clarifications sought during the session.
This specialized AI capability is engineered to support knowledge workers in demanding fields like science, finance, and policy, where rigor, comprehensive documentation, and traceability are paramount. The underlying model boasts a substantial 200,000-token context length and supports up to 100,000 completion tokens, enabling it to process vast amounts of information, reason step-by-step, and dynamically adjust its approach as new insights emerge.
Performance testing has shown impressive results, with the model achieving 26.6 percent accuracy on Humanity’s Last Exam, a challenging benchmark covering over 100 expert domains. It also reached top scores on GAIA, a standard for web-based reasoning tasks, demonstrating proficiency in information synthesis, source interpretation, and task planning. Deep Research has already been adopted within Microsoft 365 Copilot, powering the Researcher agent for synthesizing business content.
For developers, Deep Research is available via Azure subscriptions, with support for open-source tools and Python libraries. It can be accessed through Azure AI Studio or Azure CLI, allowing for the definition of prompts, multi-step pipelines, grounding controls, precision thresholds, and output formatting. The service offers transparent consumption-based pricing:
$10 per 1 million input tokens
$40 per 1 million output tokens
$2.50 per 1 million cached input tokens
Also Read:
- PolyAI and Microsoft Accelerate Enterprise Voice AI Deployment
- OpenAI’s ChatGPT Agents Poised to Challenge Microsoft Office Productivity Dominance
This strategic move by Microsoft underscores its commitment to integrating advanced GPT capabilities into programmable and composable Azure services, enabling multi-agent orchestration and knowledge-aware AI in complex enterprise environments.
