Deep Research Max

Spec

Attribute	Value
Developer	Google DeepMind
Released	2026-04-22 (public preview, Gemini API)
Announced	2026-04-22
Context window	unknown
Pricing	unknown
License	unknown
Availability	Gemini API (public preview); Google Cloud rollout planned
Foundation	Gemini 3.1 Pro
Type	Autonomous research agent

Release Date

Announced April 22, 2026 alongside Deep Research (the speed-optimized tier) and the Gemini Enterprise Agent Platform. Represents the next generation of Google's research agent product, with two-tier architecture replacing the previous single Deep Research offering.

Benchmarks

Benchmark	Score	Previous
DeepSearchQA	93.3%	66.1% (Dec 2025)
Humanity's Last Exam (HLE)	54.6%	46.4% (Dec 2025)
Source: (Google blog)

Use Cases

• Nightly due diligence reports: Asynchronous background run, exhaustive synthesis by morning
• Competitive intelligence: Web + internal data unified search
• Scientific literature review: Private dataset + open web with citation
• Financial analysis: Connects to specialized financial data sources

Two-Tier Architecture

Google released two variants simultaneously:

Feature	Deep Research	Deep Research Max
Optimization	Speed / low latency	Comprehensiveness / quality
Best for	Interactive apps, real-time	Async workflows, background jobs
Compute	Standard	Extended test-time compute
Use case	Live research UX	Overnight due diligence

Key Technical Capabilities

• MCP integration: Natively connects to private data via Model Context Protocol — internal docs, financial databases, custom datasets
• Visualization: Generates professional charts and infographics embedded in reports
• Iterative refinement: Deep Research Max reasons, searches, re-evaluates, and refines before finalizing

Compared To

Agent	Foundation	Private Data	Async Mode
Deep Research Max	Gemini 3.1 Pro	✅ via MCP	✅ native
Perplexity Deep Research	undisclosed	❌	partial
OpenAI Deep Research	GPT-5 class	❌ (web only)	❌

Sources

• Deep Research Max blog post (Google)
• VentureBeat coverage
• source snapshot