Google Gemini 3.1 Pro

ClosedMultimodal

Google

Google's frontier multimodal model with #1 GPQA Diamond score, native video and audio input, 1M context window, and fast inference at competitive pricing.

Overview

Gemini 3.1 Pro, released on February 19, 2026, is Google DeepMind's frontier multimodal model and the first to claim the #1 spot on GPQA Diamond at 94.3% — the hardest graduate-level science reasoning benchmark available. Unlike models that process different modalities through separate encoders stitched together, Gemini 3.1 Pro handles text, images, audio, and video natively within a unified architecture, with a 1M token context window and 65K output token limit.

The model introduces configurable thinking levels (Low, Medium, High) that let developers control the depth of reasoning per request. At High thinking, it tackles complex multi-step problems with extended chain-of-thought; at Low, it responds quickly for simpler queries. This flexibility is particularly valuable for production systems where latency budgets vary across different use cases within the same application.

What sets Gemini 3.1 Pro apart from competitors is its native video and audio input processing. You can feed it raw video files and audio streams directly — no preprocessing, no transcription step, no frame extraction. This makes it uniquely capable for tasks like analyzing meeting recordings, understanding video content, or processing podcast audio where other models require separate toolchains. It achieved 92.6% on MMMLU (multilingual understanding) and 80.6% on SWE-bench Verified, proving it competes at the frontier across domains.

Google offers a free tier on Google AI Studio for experimentation, with expected GA pricing around $1.50/$10 per million tokens (doubling to $4/$18 for extended context beyond 200K). The 119 tok/s inference speed and strong Arena ELO of 1493 make it a serious contender for teams already in the Google Cloud ecosystem or those who need true multimodal input processing.

Release Date

2026-02-19

Parameters

Unknown (MoE)

Context Window

1.0M tokens

Input Price

$2 / 1M tokens

Output Price

$12 / 1M tokens

Speed

119 tokens/sec

Benchmarks

BenchmarkScoreMax
GPQA Diamond94.3%100%
AA Index57%100%
Arena ELO14932000
SWE-bench Verified80.6%100%
MMMLU92.6%100%
HLE51.4%100%

Capabilities

Science Reasoning

#1 GPQA Diamond at 94.3%, the strongest graduate-level science reasoning among all models

Native Video & Audio

Process raw video files and audio streams directly without preprocessing or transcription

Configurable Thinking

Three levels (Low/Medium/High) to balance reasoning depth against latency per request

Long Context

1M token context window with 65K output limit for comprehensive document analysis

Multilingual Understanding

MMMLU 92.6%, strong performance across languages and cultural contexts

Software Engineering

SWE-bench Verified 80.6%, competitive with frontier coding models

Getting Started

1

Try for Free

Open Google AI Studio (aistudio.google.com) and select Gemini 3.1 Pro — no API key needed for the free tier

2

Get API Access

Create a project in Google AI Studio or Google Cloud, then generate an API key

3

Install the SDK

Run `pip install google-genai` (Python) or `npm install @google/genai` (Node.js)

4

Make Your First Call

Use `client.models.generate_content(model="gemini-3.1-pro")` — pass video/audio files directly as input parts

Pros & Cons

Strengths

  • +#1 GPQA Diamond (94.3%) — best science reasoning
  • +Tied #1 AA Intelligence Index (57)
  • +1M token context window
  • +Fast inference (119 tok/s)
  • +Supports video and audio input natively
  • +Free tier on Google AI Studio

Weaknesses

  • -Still in preview (not GA)
  • -Extended context pricing doubles ($4/$18)
  • -Closed source
  • -Pricing may change at GA

Best For

Science reasoningLong-context analysisMultimodal tasks (video + audio)Cost-effective frontier API

Compare

Official Links