Gemini 3.1 Pro Preview is Google's Pro-level preview model for complex reasoning, software engineering, and multi-step tasks. It improves thinking, Token efficiency, and factual consistency on the Gemini 3 Pro series foundation, making it suitable for analyzing code, long-form materials, and visual content together. On this platform, you can integrate applications through the image-and-text conversation entry point, or continuously advance tasks through the session entry point.
Clarify capacity, input and output, and calling methods before selecting a model.
Native capacity and modalities are model specifications; actual submission methods, tool execution, and session management vary by the selected entry point.
Learn what gemini-3.1-pro-preview can bring to your work.
This model is optimized for software engineering behavior and multi-step execution, making it suitable for analyzing problems using requirements, code snippets, and error logs rather than merely completing a function. Have it first organize dependencies and constraints, then propose changes, explain the scope of impact, and compile test items that need verification, producing deliverables that are easy for engineers to review.
Long-input capability is suitable for jointly reading multiple modules, design documents, and discussion records, comparing conditions and conclusions across different passages around the same issue. Combined with image understanding, it can also incorporate screenshots, charts, or interface states into the analysis. When asking questions, clearly specifying the order of materials, objects of focus, and output structure is more likely to produce useful results than broadly asking for a summary.
Native function calling and structured output are suitable for connecting reasoning results to business workflows: for example, classifying issues, extracting to-dos, or proposing the next tool request. Chat Completions provides tools and response_format configuration; actual tool execution is handled by the application, and the model continues responding based on execution results, making it easy to control each action.
Start with specific tasks to find where the model can be effective.
Provide relevant code, error stacks, expected behavior, and recent changes, and have the model analyze possible failure paths and deliver fix recommendations and a regression test checklist. When cross-file dependencies are involved, you can provide interface definitions and call sites together, and require it to distinguish between observed issues and hypotheses to be verified, avoiding only superficial changes.
Submit business questions together with image content blocks to interpret report screenshots, explain interface flows, or compare design drafts. Video understanding can accept input through video links in the Gemini conversation interface; request a content summary, key events, and details that need review. The deliverable is textual analysis, not newly generated video.
In AI Chat v2, enter research questions and file links, organize the materials first, then use the session id to follow up on differences, add conditions, or adjust the report structure. Files are submitted through the message file_url block, and tool access must comply with connection authorization. This is suitable for gradually producing document summaries, solution comparison tables, and follow-up action lists.
Choose based on task complexity, input materials, and expected results.
Compared with the Gemini 3 Pro series, 3.1 Pro Preview improves reasoning quality, Token efficiency, factual consistency, and usability for engineering tasks. If a task depends on cross-analysis of long materials, complex coding, or multi-step instructions, test it first. When migrating, compare conclusions, edit correctness, and tool requests using the same cases; do not assume a version update is necessarily better for every task.
This model is suitable for general reasoning, coding, and multimodal analysis. The official gemini-3.1-pro-preview-customtools is also available, prioritizing the use of custom tools for workflows combining bash and tools; tasks that do not depend on these tools may experience quality fluctuations. Do not use the two names interchangeably, and do not automatically choose the tool-optimized variant for ordinary analysis tasks.
From a small-scale task to production integration.
Define the goal, required inputs, and output requirements, using real business examples as a starting point.
Open the trial page, confirm the parameters supported by this entry point, then submit a small-scale task to review the results.
Keep the complete model ID, use the request format specified in the documentation, and confirm billing rules on the Pricing page.
Before production use, understand output quality and capability boundaries.
Answers to common questions about using gemini-3.1-pro-preview.
It is Google's Pro-level preview model, with the invocation ID gemini-3.1-pro-preview and no date-based fixed identifier. It is suitable for evaluating complex reasoning and engineering tasks; if your application relies on a stable output structure, it is recommended to save samples and continuously perform regression testing rather than treating Preview as a fixed snapshot.
When calling /gemini/chat/completions, write the user message content as an array containing both text and image_url blocks. image_url.url can contain a public image address or a Base64 data URI; specify the area and objective to analyze in the question, and the returned text is located in choices[0].message.content.
The model natively supports PDF understanding. In AI Chat v2, you can provide a file link through a file_url block in the message array, then use text to specify summarization or comparison requirements. Chat Completions should not directly copy this file block format; you can also extract the document text first and then submit it as message content.
In standard Chat Completions, the model can propose tool_calls, but the application needs to execute the corresponding functions and return the tool results. AI Chat v2 provides a conversation tool workflow, while access to connected resources remains subject to authorization constraints. Having the model write code and having the environment run code are separate steps; test results should come from actual execution.
When you need to control history and response formats yourself, use /gemini/chat/completions and include prior context in messages. When you want to save conversations and continue tasks using an id, use /aichat2/conversations; after enabling stateful, read answer and id, then include the same id in subsequent follow-up questions.
Model information · Updated: 2026-10-01. For invocation parameters and billing rules, see the API and pricing sections.
Start with clear goals and use real results to determine whether it is right for your work.
Input
$0.40per 1M tokens
Official $2.00
Output
$2.40per 1M tokens
Official $12.00
You pay$13,584for 142,800 credits · $0.0951 / credits46% OFF
You're billed in credits. Pick a top-up tier to see its per-credit rate — the larger the top-up, the cheaper each credit.
This is only a basic call example. See the full docs for more parameters and advanced usage.
View Docs