LLMAnonymized

Gemini 3.1 Pro Preview

Google’s 1M-context multimodal reasoning model with native tool use, vision, and web search for agentic coding and complex analysis.

Get API key
Provider
Google DeepMind
Price
$2.50 in · $15 out / 1M
Context window
1M tokens
Released
February 19, 2026
License
Proprietary

What is Gemini 3.1 Pro Preview?

Gemini 3.1 Pro Preview is Google's advanced multimodal reasoning model, released in February 2026. It processes text, images, audio, and video across a 1M-token context, offering tool use, web search, structured output, and agentic coding capabilities for complex problem-solving.

Use it privately on Venice

On Venice, Gemini 3.1 Pro Preview runs under an anonymized privacy tier — your prompts are not stored, profiled, or used for training. You get the full model with vision, reasoning, tool use, and web search without linking usage to a Google account. Note that it does not run inside a TEE or use end-to-end encryption, so Google’s infrastructure terms apply to prompt content in transit.

Anonymized
No prompt training
TEE · hardware enclave
End-to-end encrypted

What can it do?

Strengths
  • 1M-token context window ingests entire code repos, video libraries, or document sets for single-pass reasoning and synthesis.
  • Native multimodal inputs — text, images, audio, and video — with advanced reasoning and structured JSON output.
  • Built-in tool use and function calling for agentic workflows that interact with external APIs and code.
  • Web search grounding retrieves fresh information to complement the model's training knowledge.
  • Handles multiple image inputs in a single prompt for comparative visual analysis.
Limitations
  • Closed and proprietary — no open weights, so self-hosting or fine-tuning is impossible.
  • Preview release status means weights and API behavior may shift before general availability.
  • Higher per-token cost than many rivals, especially for high-volume output generation.
  • Not uncensored — Google’s safety filters apply, which can refuse certain prompts.
  • Long-context tasks can become expensive if prompts are not carefully scoped.

Gemini 3.1 Pro Preview capabilities

How to use it via API

Venice exposes an OpenAI-compatible API. Swap your base URL and call gemini-3-1-pro-preview.

curl https://api.venice.ai/api/v1/chat/completions \
  -H "Authorization: Bearer $VENICE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gemini-3-1-pro-preview",
    "messages": [{ "role": "user", "content": "Explain quantum tunneling simply." }]
  }'

Specifications

MakerGoogle DeepMind
ReleasedFebruary 19, 2026
ModalityMultimodal (text, image, audio, video in; text out)
Knowledge cutoffJanuary 2025
ArchitectureBased on Gemini 3 Pro; detailed specs undisclosed
ParametersNot disclosed
Open weightsNo — proprietary
Context window1,000K tokens
Max output32.768K tokens
CapabilitiesVision, Function calling, Reasoning, Web search
Privacy on VeniceAnonymized — prompts not stored
Available on Venice sinceFeb 2026

Pricing

Billed per token on Venice: $2.50 per 1M input tokens and $15 per 1M output tokens.

Input / 1M tokens
$2.50
Output / 1M tokens
$15
Cached input / 1M
$0.50

New Venice accounts include a free daily allowance and 500 welcome credits — no credit card required.

Gemini 3.1 Pro Preview vs alternatives

ModelContext windowPrice (Venice)Open weightsKey strength
Gemini 3.1 Pro Preview1M tokens$2.50 in · $15 out / 1MNoMultimodal reasoning & agentic coding
Claude Opus 4.81M tokens$6 in · $30 out / 1MNoDeep reasoning & coding
Grok 4.31M tokens$1.42 in · $2.83 out / 1MNoSearch & general chat
DeepSeek V3.2160K tokens$0.33 in · $0.48 out / 1MYesOpen-weight efficiency

Google’s top-tier 1M-context model with vision, reasoning, and tool use for agentic workflows.

What is it good for?

  • Repo-scale software engineering — reading large codebases, planning refactors, and executing tests via tool use.
  • Long-context RAG over massive document sets, transcripts, and media for legal, financial, or research analysis.
  • Multimodal content extraction and moderation from images, audio, and video with structured JSON output.
  • Agentic workflow automation with function calling and web search for autonomous research and task completion.
  • Complex data synthesis and planning across long-form reports and spreadsheets.

Prompting tips

  • Chunk long inputs with clear headers — the 1M context is powerful but benefits from explicit structure.
  • Use structured JSON schemas for reliable data extraction from multimodal inputs.
  • Define tool schemas and boundaries tightly to avoid expensive recursive loops in agentic workflows.
  • For vision tasks, provide multiple images in a single prompt to exploit the native multimodal context.
  • Scope long-context prompts carefully; unused tokens still count toward input pricing.

Version history

Gemini 3.1 Pro Preview
2026-02

CurrentCurrent preview release with improved reasoning and agentic capabilities.

Frequently asked questions

Gemini 3.1 Pro Preview is Google's advanced multimodal reasoning model, released in February 2026. It processes text, images, audio, and video across a 1M-token context, with built-in tool use, web search, and structured output for complex agentic tasks.

$2.50 per 1M input tokens and $15 per 1M output tokens. Cached input is billed at $0.50 per 1M tokens. You pay only for what you use with no subscription required.

It is not free or open source. It is a closed, proprietary Google model. You access it through Venice's API with per-token pricing. For open-weight alternatives, try DeepSeek V3.2 or Kimi K2.6 on Venice.

Yes. Venice exposes its function-calling and structured-output capabilities, so you can connect it to external APIs, code interpreters, and search tools for agentic workflows.

Choose Gemini 3.1 Pro for multimodal inputs, 1M-context RAG, and agentic coding with web search. Choose Claude Opus 4.8 if you prefer Anthropic’s reasoning style and are willing to pay a higher per-token rate for deep textual analysis.

Venice runs it under an anonymized privacy tier — your prompts are not stored, profiled, or used for training. However, it does not run in a TEE or use end-to-end encryption, so Google may see prompt content under their own terms.

Ingesting entire code repositories, long video transcripts, large document sets, or multi-source research briefs in a single prompt, then reasoning across them to generate structured plans, summaries, or code changes.

Yes. On Venice it accepts audio and video inputs alongside images, enabling multimodal reasoning across media types in a single prompt.

Related models

Run Gemini 3.1 Pro Preview privately.

No prompt logging. No data used for training. Free to start — no credit card.

Room