Deep Research Max, announced by Google DeepMind on April 22, is an enterprise-grade autonomous research agent powered by Gemini 3.1 Pro. The system is designed for multi-source research synthesis, pulling from public web sources and private enterprise data simultaneously, then producing structured research outputs with reasoning-chain verification. That’s Google DeepMind’s characterization. The independent evaluation picture is thinner: eval status is currently pending.
The headline enterprise feature, reported by SiliconAngle (primary source confirmation against Google DeepMind’s official announcement is pending), is native MCP support for FactSet and PitchBook. Both are core financial data platforms. If the integration holds at the depth implied, Deep Research Max becomes meaningfully different from a general web research agent, it can reach licensed, structured financial data that public web search can’t touch.
On benchmarks, Google DeepMind reports Gemini 3.1 Pro scored 77.1% on ARC-AGI-2, per reporting by VentureBeat. Independent evaluation of this figure is pending. On the DeepSearchQA benchmark, Google DeepMind’s internal research evaluation, the company reports 93.3%. That’s a self-reported score on an internal benchmark, which means methodology and scope aren’t independently verifiable yet. Both figures are noted here as vendor-reported, not confirmed.
The enterprise research use case is where this gets interesting. A financial analyst running a company deep-dive today typically combines Bloomberg terminal data, SEC filings, news aggregation, and manual synthesis. Deep Research Max’s implied workflow compresses that process. Whether it executes reliably at production depth, without hallucinating citations or misattributing figures, is the real test, and it’s one that requires independent evaluation to answer.
The timing of this release alongside OpenAI’s GPT-5.5 Pro and Workspace Agents is notable. Both companies are moving toward enterprise agentic automation in the same week. But their surface areas differ. Deep Research Max is targeting a specific, high-value research workflow, financial analysis and synthesis. Workspace Agents is broader enterprise automation. These are different buyer conversations, even if the underlying technology overlaps.
What to watch: Primary source confirmation of the FactSet and PitchBook integrations matters before enterprise procurement decisions move forward. SiliconAngle’s reporting should trace to a Google DeepMind announcement or a statement from FactSet or PitchBook directly. If confirmed, this integration makes Deep Research Max a serious tool for financial services AI teams, not a general-purpose assistant in a vertical wrapper. Independent ARC-AGI-2 evaluation from Epoch AI or a comparable assessor would also clarify how the benchmark figures hold up outside vendor testing conditions.
TJS synthesis: Deep Research Max is a targeted enterprise bet, not a broad-market play. Google DeepMind is identifying a specific high-value workflow, financial research synthesis, and building direct integrations with the data infrastructure those workflows depend on. That’s a more defensible enterprise strategy than general-purpose automation: if the FactSet and PitchBook integrations work well at depth, the switching cost for a financial services firm that embeds Deep Research Max rises quickly. The verification gap on both the integrations and the benchmarks is the practical barrier to enterprise adoption right now. Those gaps close with confirmed source data and independent evaluation.