Google Ships Deep Research Max with MCP and $4.80 Per-Report Pricing
Google launched Deep Research and Deep Research Max on April 21 — two autonomous research agents built on Gemini 3.1 Pro with identical benchmark scores (DeepSearchQA 93.3%, HLE 54.6%). The 3.9× price gap ($1.22 vs $4.80/report) has an empirical answer: an independent seven-task benchmark found the cheaper standard tier won or tied five of seven. Three independent sources covered the launch in the same 24-hour window, confirming it as the week's clearest enterprise-AI pricing signal.
What the Source Actually Says
Chew Loong Nian's Level Up Coding benchmark ran both agents through seven tasks — product comparison, regulatory summary, literature review, competitive landscape with private documents, earnings-call digest, tech-news briefing, and an industry writeup — using identical prompts, MCP servers, and success criteria. Standard Deep Research ($1.22, 250K input + 60K output tokens) won or tied five. Deep Research Max ($4.80, 900K input + 80K output) earned its 3.9× premium on only two task shapes — details are paywalled, but the framing points to tasks requiring deep synthesis across private data sources.
The MCP integration is the architectural pivot both tiers share. Coverage from @yaelkroy on X noted that MCP support lets both agents connect to internal databases, corporate knowledge stores, private document repositories, and specialized financial data — not just the public web — producing "analyst-grade work with provenance and verifiable citations." Google positioned Max as searching deeper and reasoning longer, with its 900K input context (versus 250K for standard) as the mechanical basis for that claim.
The NLP Newsletter's AI Agents Weekly placed Deep Research Max alongside GPT-5.5 and DeepSeek V4 as among the week's defining signals in an "agents are now product" moment — editorial confirmation that autonomous research agents have moved from research curiosity to evaluable enterprise tools. Both tiers also support native chart generation and private-data integration through Google's new Interactions API.
Strategic Take
Standard tier wins five of seven benchmark tasks at 3.9× lower cost. Max earns its premium specifically on tasks requiring deep synthesis across private data — but since MCP access is shared across both tiers, the structural unlock is already present in the cheaper option. Run your own task benchmark before committing $4.80/report at scale.