Google Gemini Pro

Gemini Pro is the frontier tier of Google's Gemini family, aimed at the hardest reasoning, research and agentic-coding work rather than high-volume throughput. It is bought per token through the Gemini API, Google AI Studio and the Gemini Enterprise Agent Platform, and it removes the ceiling that cheaper tiers hit on multi-step problems and deep multimodal analysis.

Gemini Pro is the top capability tier of Google's Gemini model family, positioned above Flash and Flash-Lite for complex reasoning, multimodal understanding and agentic coding. As of August 2026 the line is in an unusual state: Gemini 3.1 Pro, released as a preview on 19 February 2026, is the newest Pro model and remains labelled Preview rather than generally available; Gemini 3 Pro has been deprecated and shut down; and Gemini 2.5 Pro is the stable, production-labelled option teams can build against with a GA commitment. Gemini 3.1 Pro accepts text, images, video, audio and code across a 1M-token input window and returns up to 64k output tokens, and Google reports 44.4% on Humanity's Last Exam, 94.3% on GPQA Diamond and 80.6% on SWE-Bench Verified for agentic coding. It is reachable through the Gemini app, Google AI Studio, the Gemini API, the Gemini Enterprise Agent Platform and Google Antigravity. Pricing is consumption-based and tiered by prompt length: $2.00 per million input tokens and $12.00 per million output tokens for prompts at or under 200k tokens, rising to $4.00 and $18.00 above that threshold, with batch mode at half rate and no free tier at all. Gemini 2.5 Pro remains cheaper at $1.25 and $10.00 per million tokens under 200k. Google has said a Gemini 3.5 Pro is in limited partner testing after slipping past the June 2026 target it set at I/O, which leaves enterprise buyers choosing between a preview-status frontier model and a stable but older one.

Category

AI Models & APIs

Pricing

Usage-based

Target Market

CTOs, CIOs, Chief Data Officers, Enterprise Developers, Data Scientists, AI Research Leads

Deployment

Cloud-only, API-based

Founded

1998

Headquarters

Mountain View, California, United States

Team Size

500+

• ✓
  1M-token input window with 64k output

  Holds whole codebases, long video or large document sets in context, so deep analysis does not depend on a retrieval layer's recall.

• ✓
  Frontier reasoning benchmarks

  Google reports 44.4% on Humanity's Last Exam and 94.3% on GPQA Diamond, the evaluations that separate frontier models from mid-tier ones.

• ✓
  Agentic coding capability

  Scores 80.6% on SWE-Bench Verified and is positioned explicitly for agentic and vibe coding with improved multi-step tool use.

• ✓
  Native multimodal understanding

  Processes text, images, video, audio and code together, which matters for review workflows where the evidence is not all text.

• ✓
  Long-context pricing tier

  Prompts above 200k tokens are billed at a distinct higher rate, so cost is predictable but doubles once you cross that boundary.

• ✓
  Batch processing at half price

  Asynchronous batch runs cost $1.00 input and $6.00 output per million tokens under 200k, making offline analysis materially cheaper.

• ✓
  Enterprise deployment surface

  Available on the Gemini Enterprise Agent Platform with regional ML processing, configurable zero data retention and a self-serve HIPAA BAA.

• •
  Repository-scale engineering agents

  Plan and execute multi-file changes across a large codebase where the model must hold architecture and tests in context simultaneously.

• •
  Scientific and technical literature analysis

  Work through dense research material at graduate level, where GPQA Diamond-class reasoning separates a usable answer from a plausible one.

• •
  Long-form video and evidence review

  Analyse hours of video or audio alongside supporting documents in one prompt, producing structured findings rather than a transcript.

• •
  Offline batch analysis of large corpora

  Run deep synthesis across contracts, filings or incident reports overnight at half the standard token rate via batch mode.

• •
  Escalation tier behind a cheaper model

  Route only the requests a Flash-tier model fails to Gemini Pro, keeping frontier cost confined to the small share of hard cases.

Consumption-based with no seat licences and no free tier on any Pro model, which is the single biggest difference from the Flash line. Gemini 3.1 Pro Preview lists at $2.00 input and $12.00 output per million tokens for prompts up to 200k, then $4.00 and $18.00 beyond it — and the long-context surcharge applies to the whole request, so crossing 200k roughly doubles the bill rather than just pricing the excess. Batch halves every rate, context caching costs $0.20 per million plus $4.50 per million tokens per hour of storage, and the older but stable Gemini 2.5 Pro is meaningfully cheaper at $1.25/$10.00. Search grounding is billed on top at $14 per 1,000 queries after 5,000 free monthly requests.

This page was written from 8 sources, 6 on domains other than ai.google.dev.

1. 1.ai.google.dev — pricingvendor
2. 2.deepmind.google — pro
3. 3.ai.google.dev — modelsvendor
4. 4.cloud.google.com — pricing
5. 5.techcrunch.com — googles new gemini pro model has record benchmark scores aga
6. 6.hn.algolia.com — hn.algolia.com
7. 7.unite.ai — google ships three gemini flash models as its flagship slips
8. 8.aiprovidertrust.com — gemini vertex