Google Gemini 3.5 Pro Launches With 2-Million Token Context

30-Second Summary

Google DeepMind launched Gemini 3.5 Pro on 17 July 2026, delivering a 2-million-token context window and a rebuilt reasoning architecture. The model arrives six weeks late, having been rebuilt from scratch after engineers identified failures in recursive tool-calling in the original version. It is now available via the Gemini API and Vertex AI, making it accessible to businesses of any size. Deep Think, the model's extended reasoning mode, is available at the $250 per month Ultra subscription tier.

At a Glance

• Topic: Model Releases
• Company: Google DeepMind
• Date: 17 July 2026
• Announcement: General availability of Gemini 3.5 Pro via the Gemini API and Vertex AI
• What Changed: The frontier context window doubled to 2 million tokens, the largest available at the frontier
• Why It Matters: Businesses can now process entire knowledge bases, document archives, and codebases in a single AI request
• Who Should Care: Any operator using AI for document analysis, customer support, coding, or knowledge management

Key Facts

• Company: Google DeepMind
• Launch Date: 17 July 2026
• What Changed: New pretraining run, 2-million-token context window, Deep Think reasoning layer
• Who It Affects: Business operators using the Gemini API, Google Workspace, or Vertex AI
• Primary Source: Google AI blog; Google DeepMind official announcement

What Happened

Google DeepMind launched Gemini 3.5 Pro on 17 July 2026, marking the general availability of its most capable model to date. The model was originally announced at Google I/O in May 2026 with a June target, but the company delayed it by six weeks after engineers discovered structural failures in the original model's recursive tool-calling behaviour. Rather than patch the existing model, Google DeepMind rebuilt it on an entirely new pretraining run.

The headline specification is a 2-million-token context window, double anything currently available at the frontier. In practical terms, this allows a single API request to include two million words of text, code, or mixed data. Users can pass entire books, codebases, or research archives in a single session. The model also introduces Deep Think, an extended reasoning mode designed for complex multi-step tasks including mathematical reasoning, legal analysis, and long-horizon planning. Deep Think is available exclusively to users on the Gemini Ultra subscription tier at $250 per month.

Gemini 3.5 Pro is available via the public Gemini API and through Vertex AI, Google Cloud's enterprise AI platform. The Vertex AI route provides private deployment options, data residency controls, and enterprise service level agreements not available on the consumer product. Leaked pricing information circulating before launch suggested rates near $1.25 per million input tokens and $10 per million output tokens for the standard tier, though Google has not published official pricing figures.

The launch coincides with the opening of the 2026 World Artificial Intelligence Conference in Shanghai, where Chinese President Xi Jinping is attending in person for the first time since the event began in 2018. The convergence underscores that AI competition has become a top-tier strategic priority for both the United States and China.

Why It Matters

• The 2-million-token context window removes the primary constraint on how much context a business can feed an AI model in a single session. Entire contracts, support archives, and company knowledge bases are now processable in one pass, without chunking or summarisation workarounds.
• Gemini 3.5 Pro arrived five days after GPT-5.6 and nine days after Grok 4.5, meaning the frontier model landscape has shifted significantly in less than a fortnight. Businesses now have genuine competitive options at the top tier.
• The Vertex AI deployment path matters for operators in regulated industries. Private deployment combined with enterprise SLAs addresses the data governance objections that have slowed AI adoption in legal, finance, and healthcare firms.
• The Deep Think reasoning mode raises the ceiling on what an AI model can reliably deliver for complex analytical tasks, moving the capability closer to senior professional grade for tasks requiring sustained multi-step logic.
• Autonomous workflow capabilities built into the model, designed to manage multi-step coding and tool execution with minimal human oversight, are relevant to businesses looking to automate processes that currently require a person to coordinate multiple software tools.

The David and Goliath View

The arrival of Gemini 3.5 Pro is significant not just for its specifications but for what it signals about the pace of change. Three frontier models, GPT-5.6, Grok 4.5, and now Gemini 3.5 Pro, launched within a fortnight. For businesses that have been waiting for the market to settle before committing to an AI stack, the message is that it will not settle. The competition is accelerating, not slowing.

The 2-million-token context window is the most immediately applicable capability for lean organisations. A business with 50 employees likely has more institutional knowledge locked in documents, emails, and transcripts than any individual staff member can hold in memory. A model that can read and reason across that entire archive in real time is, in effect, a new kind of staff member who has already been fully onboarded. That capability is now available via an API for a few dollars per call.

The practical recommendation is to move from evaluating AI to deploying it against a specific, measurable workflow this month. The cost of waiting is now higher than the cost of a wrong first choice. Pick your highest-volume document-heavy process, test Gemini 3.5 Pro's context window against it in a Vertex AI sandbox, and measure the time saved. That evidence is more valuable than any benchmark score.

Where This Fits in the AI Stack

AI Growth Engine: The 2-million-token context window enables AI-powered analysis of full customer histories, sales archives, and market research sets in a single request, directly supporting more informed and faster business decisions.

Employee Amplification Systems: Autonomous workflow capabilities and the Deep Think reasoning layer allow staff to delegate multi-step analytical and operational tasks to AI, reducing time spent on research, synthesis, and coordination.

Secure AI Brain: Vertex AI deployment provides the private infrastructure, data residency controls, and enterprise SLAs required to build a company knowledge layer on top of Gemini 3.5 Pro without exposing sensitive data to public endpoints.

Questions Operators Are Asking

What can I actually do with a 2-million-token context window? You can feed the model an entire year of customer support tickets and ask it to identify the top recurring issues. You can load a full set of supplier contracts and ask it to flag non-standard clauses. You can pass a complete codebase and ask it to document every function. Tasks that previously required staff to manually summarise and distil large document sets can now be handled in a single API call.

Is Gemini 3.5 Pro available in Google Workspace? Google integrates its models progressively into Workspace products including Docs, Sheets, and Gmail. Enterprise Workspace customers on Gemini for Workspace plans should expect access through the Vertex AI route initially, with broader Workspace integration following over subsequent weeks. Check your Google Workspace admin console for current availability.

Should I switch from GPT-5.6 or Claude Sonnet 5 to Gemini 3.5 Pro? All three are frontier-class models and the right choice depends on your specific workflow. Gemini 3.5 Pro's 2-million-token context window is a genuine differentiator for document-heavy tasks. GPT-5.6 and Claude Sonnet 5 may perform better on other tasks. Running your actual use case against each model is more reliable than relying on published benchmarks.

What does Deep Think cost and is it worth it? Deep Think requires the Gemini Ultra subscription at $250 per month. For most business operators, the standard Gemini 3.5 Pro tier will handle the majority of tasks. Deep Think is worth evaluating for workflows requiring extended reasoning, such as complex financial modelling, multi-document legal analysis, or detailed technical planning where quality of reasoning matters more than speed or cost.

Is Vertex AI the right deployment path for a company our size? Vertex AI is worth considering for any business handling commercially sensitive data, data covered by regulatory requirements, or data subject to client confidentiality obligations. The enterprise tier provides data residency controls and service level agreements. For businesses without these constraints, the public Gemini API is simpler and less expensive to start with.

Citable Summary

What happened: Google DeepMind launched Gemini 3.5 Pro on 17 July 2026, delivering a 2-million-token context window and a Deep Think extended reasoning mode via the Gemini API and Vertex AI.

Why it matters: The 2-million-token context window doubles the frontier and allows businesses to process entire knowledge bases, document archives, and codebases in a single AI request, removing the primary constraint on AI-powered analysis at scale.

David and Goliath view: Three frontier models launched in a fortnight. The cost of waiting now exceeds the cost of a wrong first deployment. Businesses should test Gemini 3.5 Pro against their highest-volume document workflow this month and measure the result.

Offer relevance:

• AI Growth Engine: Full-archive analysis capability enables richer, faster business intelligence.
• Employee Amplification Systems: Autonomous workflows and Deep Think reasoning reduce time spent on research and multi-step tasks.
• Secure AI Brain: Vertex AI enterprise deployment supports private, compliant AI infrastructure for regulated or sensitive environments.