Gemini 3.1 Pro Preview
Neural Network
Gemini 3.1 Pro Preview is Google's reasoning model for code, agent tasks, and working with multimodal data.
Specifications
- Input
- text, images, documents
- Output
- text
- Context size
- 1 048 576 tokens
- Max answer length
- 65 536 tokens
- Reasoning effort
- low, medium, high
- Capabilities
- reasoning, function calling
Prompt cost
(per 1M tokens)
Answer cost
(per 1M tokens)
Image prompt
(per 1K tokens)
How it works Gemini 3.1 Pro Preview?
Gemini 3.1 Pro Preview is a Google reasoning model built on the Gemini line's multimodal foundation. Compared to the previous version, it handles long agent chains more stably, excels in software engineering, and uses tokens more efficiently in complex multi-step scenarios. The batch variant is designed for bulk task submission when processing volume matters more than instant response. BotHub provides access without VPN or foreign cards: pay in rubles with a Russian card only for tokens used, and unused balances do not expire. With over 250 models in one window, you can compare responses or switch without rewriting integrations—a single OpenAI-compatible API handles them uniformly. Chats are encrypted via AES-GCM, and companies get contracts, invoices, EDI, and an admin panel with limits. Developers can use it for refactoring and debugging, analysts for insights from diverse data, product teams for agents and automation, and researchers for STEM tasks and external function calling.Frequently asked questions about Gemini 3.1 Pro Preview
You can use generated results for commercial purposes. You own all rights to the content you create. The only restriction: make sure your prompt does not include copyrighted third-party material. You are responsible for respecting the rights to any input data.
Gemini 3.1 Pro Preview in batch mode works with text, documents, and images: it analyzes long materials, writes and edits code, and solves reasoning tasks. A context window of up to 1,048,576 tokens allows you to hold an entire project, contract, or large log dump in a single request.
Up to 65,536 tokens in a single response—this is dozens of pages or a large code module in its entirety, without manual stitching. If the task is even larger, break it into steps: tokens in BotHub do not expire, you pay only for what you use.
Yes, reasoning before answering is confirmed: the model first builds a chain of steps and only then provides the result. This is noticeable in math, logic, code analysis, and multi-step instructions—the answer is more accurate than with immediate generation. You pay for the length of reasoning in tokens, without a subscription.
Yes, both. The model calls your functions with arguments and returns structured JSON according to a specified schema—this is convenient for agents, document parsing, and integrations. Connection is via the unified OpenAI-compatible BotHub API, so switching models does not require rewriting code.
Images and documents—yes, these are confirmed input types: screenshots, diagrams, scans, reports, and dumps can be sent with your question. There is no information about audio and video in our data, so please check the interface—available formats are visible when uploading a file.
There are no specific notes about languages in our data, so we cannot promise the accuracy of formulations—it is more reliable to test it with your own task. Open the model on the website or in the Telegram bot without a VPN or foreign card, enter a typical query, and compare it with another model in the same window.