Gemini 3.1 Pro: the strongest reasoning in the Gemini line

Gemini 3.1 Pro is Google's top Pro model of the Gemini 3 generation. It is the strongest model in the Gemini line for reasoning: it handles complex analytical tasks, writes and reviews code, and works with images and documents. On Genosai it runs with a 120,000-token context, accepts images and supports function calling. The price is 160 / 960 ₽ per 1M tokens (input / output), with no subscription and no VPN.

Updated: September 27, 2026

Contents

What is Gemini 3.1 Pro

Gemini 3.1 Pro is Google's senior model in the Gemini 3 generation. If you need the Gemini model with the strongest reasoning, this is it. Next to it in the Genosai catalog are lighter models of the same generation: the fast Gemini 3 Flash, the most affordable Gemini 3.1 Flash Lite and the newer fast Gemini 3.8 Flash. The previous Pro model is Gemini 2.5 Pro.

The difference between Flash and Pro is purpose. Flash models are built for speed and volume: classification, short copy, quick chat replies. Pro is for tasks where a mistake costs more than the tokens: breaking down numbers and causes, code review, analyzing a document dozens of pages long, a project plan with dependencies. Here the model has to keep many conditions in mind, go through several reasoning steps and explain its conclusion rather than just produce plausible text.

Gemini 3.1 Pro works with text and images. You can feed it a UI screenshot, a photo of a document, a diagram or a chart, and it will analyze the content. The answer is always text. Function calling is supported, so the model works well as the brain of agents and automations.

On Genosai, Gemini 3.1 Pro runs with a 120,000-token context. The client price is 160 / 960 ₽ per 1M tokens (input / output). It works from Russia without a VPN, you can pay with a Russian bank card, and no subscription is required.

Capabilities

Choose Gemini 3.1 Pro when the task is complex and something depends on the quality of the answer. Here are the main scenarios.

Analysis and reasoning

The model breaks down numbers, looks for causes and links, compares options and explains how it reached a conclusion. That is useful for sales reports, unit economics, funnel analysis and testing hypotheses. Asking it to show the math works well: the model writes formulas and intermediate values, which are easy to check.

Code and code review

Gemini 3.1 Pro writes code, explains someone else's, finds bugs and weak spots, and proposes refactoring. A 120,000-token context fits a large module together with the task description and an error log. For tricky logic, raise reasoning effort to medium or high.

Images and documents

A screenshot with an error, a photo of an invoice, a process diagram, a chart from a slide deck: all of these can be attached to a request. The model describes what is in the image, extracts data, finds inconsistencies or suggests what to improve in an interface.

Structured output and agents

The model keeps the format you set: a table, a numbered plan, JSON with the fields you need. Together with function calling, this lets you embed Gemini 3.1 Pro in automations such as parsing inbound leads, filling CRM fields or drafting replies, and pass the result on without manual edits.

Examples prompt and response

Below are real Gemini 3.1 Pro answers produced through the Genosai API with reasoning effort set to medium. The first example is an analytical business task, the second is a code review, the third is a structured plan. Answers are quoted verbatim; the long table is trimmed with "…".

Prompt: An online store: in August 1,200 orders at an average order value of $45 and a 2% conversion rate; in September 1,050 orders, AOV $47, conversion 1.6%. Traffic grew. Explain what happened to revenue, name 3 likely causes of the conversion drop and what to check first. Be concise, show the math.

Gemini 3.1 Pro answer:
### Revenue & Traffic Math
**Revenue dropped by $4,650 (-8.6%).** Even though Average Order Value (AOV) and traffic increased, the drop in conversion rate resulted in fewer total orders.

*   **August:**
    *   Revenue: 1,200 orders × $45 AOV = **$54,000**
    *   Traffic: 1,200 orders / 0.02 conversion = **60,000 visitors**
*   **September:**
    *   Revenue: 1,050 orders × $47 AOV = **$49,350**
    *   Traffic: 1,050 orders / 0.016 conversion = **65,625 visitors**

### 3 Likely Causes of the Conversion Drop
1.  **Low-Quality Traffic Surge:** A new marketing campaign, viral post, or broad ad targeting brought in "window shoppers" who inflated visitor numbers without intent to buy.
2.  **Price Increases or Fewer Discounts:** The increase in AOV ($45 to $47) suggests prices went up or a promotion ended. This higher cost may have priced out some potential buyers.
3.  **Inventory Shortages:** An influx of traffic may have bought out popular, lower-priced items early in the month, leaving later visitors with out-of-stock pages (killing conversion) and only higher-priced items left to buy (raising AOV).

### What to Check First
**Traffic sources and campaign performance.**
Go into your analytics and segment the traffic by channel/source. Look for where the extra 5,625 visitors came from. If a specific ad campaign, social channel, or affiliate drove a massive spike in traffic with a near-zero conversion rate, you have found the culprit.

_Generated on Genosai.io with Gemini 3.1 Pro._

Prompt: Review this Python function: find bugs and risks, give a fixed version. Be concise.

def get_discount(user, orders=[]):
    total = 0
    for o in orders:
        total += o['sum']
    if total > 10000:
        return total * 0.1
    elif total > 5000:
        return total * 0.05

Gemini 3.1 Pro answer:
### Bugs and Risks

1. **Mutable Default Argument:** `orders=[]` is evaluated only once at definition. If mutated later, the changes persist across function calls. Use `None` instead.
2. **Missing Return Value:** If `total <= 5000`, the function implicitly returns `None`, which will cause `TypeError`s if the caller tries to do math with the result. It should return `0`.
3. **Missing Key/Type Errors:** `o['sum']` will raise a `KeyError` if the key is missing, or a `TypeError` if `o` isn't a dictionary.
4. **Unused Variable:** The `user` parameter is never used.
5. **Floating Point Precision:** Using floats for currency/financials can lead to precision errors (e.g., `0.1 + 0.2 = 0.30000000000000004`). Consider using the `decimal` module for real-world financial apps.

### Fixed Version

def get_discount(user, orders=None):
    if orders is None:
        orders = []

    # Safely get 'sum' defaulting to 0 if missing
    total = sum(o.get('sum', 0) for o in orders)

    if total > 10000:
        return total * 0.1
    elif total > 5000:
        return total * 0.05

    return 0.0 # Explicitly return 0 if no discount applies

*(Note: If strict financial accuracy is required, replace `float` multiplication with the `decimal.Decimal` type).*

_Generated on Genosai.io with Gemini 3.1 Pro._

Prompt: Draft a 4-week launch plan for an online course at a small school: week by week, 2-3 tasks per week and one metric that shows the week went well. Format it as a table.

Gemini 3.1 Pro answer:
Here is a 4-week launch plan tailored for a small school, formatted as a table.

| Week | Phase | Key Tasks (2-3 per week) | Success Metric |
| :--- | :--- | :--- | :--- |
| **Week 1** | **Pre-Launch & Awareness** | 1. Finalize the course landing page and test the enrollment/payment system. 2. Send a "Coming Soon" teaser email to the school's existing mailing list (parents, alumni, community). 3. Post teaser graphics on the school's social media channels. | **Landing Page Traffic:** Achieving a target number of unique visitors to the new course page. |
| **Week 2** | **Official Launch & Early Bird** | 1. Officially open registration and send the launch email offering a limited-time "Early Bird" discount. 2. Host a short 15-minute live Q&A session (via Zoom, Instagram, or Facebook) to explain the course benefits and answer questions. | **Early Bird Enrollments:** Securing the first batch of sign-ups (e.g., hitting 25% of your total enrollment goal). |
…

_Generated on Genosai.io with Gemini 3.1 Pro._

We checked the math in the first answer: revenue of $54,000 and $49,350, a drop of $4,650 (8.6%), and traffic of 60,000 and 65,625 visitors are all correct. The fixed function in the second answer is correct too, though o.get('sum', 0) silently skips orders with no amount, and sometimes an explicit error is better. Figures such as "25% of your enrollment goal" in the third answer are the model's suggestions, not verified data. Each answer used between 900 and 1,650 output tokens including reasoning, roughly 0.9 to 1.6 ₽.

How to use on Genosai

  1. Open the Genosai studio on the Text tab.
  2. Pick Gemini 3.1 Pro in the model list.
  3. Open the chat settings panel: temperature, top_p, a response limit of up to 16,384 tokens, reasoning effort and a system prompt.
  4. Describe the task: the goal, the source data and the answer format. Attach images if needed.
  5. Get the answer and refine it in the same dialog; the history is kept within the context.

Billing is per token, with no subscription. You can top up with a Russian bank card, and no VPN is needed.

Reasoning effort

The chat settings panel has a Reasoning effort parameter, called reasoning_effort in the API. Gemini 3.1 Pro offers three values: low, medium and high. Low is the default. There is no "no reasoning" option for this model: it always thinks the answer through at least a little.

On simple questions reasoning costs almost nothing. In our test, a question at medium effort used about 60 reasoning tokens, roughly 0.06 ₽. For math, logic, complex code and multi-step analysis, raise the effort to medium or high: the answer takes longer but is better thought out. For short copy and quick lookups, keep it at low.

Through the API

The Genosai API follows the OpenAI format, so familiar SDKs work. The list of available models is at GET /v1/models.

curl https://api.genosai.io/v1/chat/completions \
  -H "Authorization: Bearer sdk_YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gemini-3.1-pro",
    "reasoning_effort": "medium",
    "messages": [{"role": "user", "content": "Analyze this sales report..."}]
  }'

Prompts

These templates play to the model's strengths: analysis, code and structure. Replace the descriptions in angle brackets with your own data.

Here is a sales export for two months: <table>. Find what changed, calculate each factor's contribution to revenue and suggest 3 hypotheses to test.
Review this code: find bugs, security risks and places to simplify. Give a fixed version and a test: <code>
Read the contract and list in a table: each party's obligations, deadlines, penalties and clauses worth discussing with a lawyer: <text>
Look at this account dashboard screenshot and list what keeps the user from finding the payment button. Suggest 5 changes in priority order.
Draft a 6-stage project plan: for each stage give tasks, dependencies, the main risk and a definition of done: <project description>
Parse this customer request and return JSON with the fields: name, contact, request summary, budget, urgency, next step.

Generation cost

The client price of Gemini 3.1 Pro on Genosai is 160 / 960 ₽ per 1M tokens (input / output). Only actual usage is charged: the length of the request, the attached materials and the answer.

To estimate a budget: a request with 2,000 input tokens and a 1,000-token answer costs about 1.3 ₽. Reasoning tokens are billed as output tokens. On a simple question that is a few dozen tokens, but on a hard task at high effort there can be noticeably more, so turn on high effort where it really matters.

There is no subscription, and top-ups work with Russian bank cards. Current prices are in the Pricing section.

How it compares

Gemini 3.1 Pro is the strongest model in the Gemini line and sits in the middle price tier among other providers' top models.

ModelPrice ₽ per 1M (input / output)ContextNiche
Gemini 3.1 Pro160 / 960120,000Complex analysis, code, images
Gemini 3 Flash40 / 240120,000Fast tasks in the Gemini 3 generation
Gemini 3.1 Flash Lite20 / 120120,000Cheapest 3.x model
Gemini 2.5 Pro105 / 840120,000Previous-generation Pro
GPT-6 Sol160 / 800256,000Code and automation
GPT-5.6 Sol400 / 2400256,000Previous-generation GPT flagship
Claude Sonnet 5190 / 940800,000Very large volumes of text
DeepSeek V4 Pro3.2 / 158800,000Low-cost, text only

Within the line, Gemini 3 Flash is four times cheaper and Gemini 3.1 Flash Lite eight times cheaper; for simple, high-volume tasks they are enough. Compared with Gemini 2.5 Pro, the new Pro is about 1.5 times more expensive on input and 14% more on output, but it is a newer generation.

Among competitors, GPT-6 Sol has the same input price, is 17% cheaper on output and offers a 256,000-token context. GPT-5.6 Sol costs 2.5 times more than Gemini 3.1 Pro. Claude Sonnet 5 is more expensive on input, nearly equal on output and takes an 800,000-token context. DeepSeek V4 Pro is 50 times cheaper on input and about six times cheaper on output, but it works with text only and does not accept images.

Limitations and tips

On Genosai, Gemini 3.1 Pro has no built-in web search: the model does not know the latest news, rates or prices. If the answer depends on current data, pass it in the request. Output is text only: the model analyzes images but does not draw them.

The context is 120,000 tokens. For very long materials, split them into parts or pick a model with a larger window, such as Claude Sonnet 5. The answer is capped at 16,384 tokens, so ask for long texts in parts.

Reasoning cannot be turned off, but you can control its depth. Keep low effort for simple questions and use medium or high for math, logic and code. If the task is simple and high-volume, do not overpay for Pro: Gemini 3 Flash will handle it for less.

State the format explicitly: a table, a number of points, JSON, a length limit. For code, pass related files, library versions and the error text. And check the result manually whenever money, security or important decisions depend on it: run the code and verify numbers and facts against the source. Even the strongest model can make mistakes.

FAQ

What is Gemini 3.1 Pro?

It is Google's top Pro model of the Gemini 3 generation and the strongest model in the Gemini line for reasoning. It is built for analysis, programming and work with images and documents. On Genosai it has a 120,000-token context and answers with text.

How much does Gemini 3.1 Pro cost on Genosai?

The client price is 160 / 960 ₽ per 1M tokens (input / output). There is no subscription: you pay only for the actual size of the request and the response. Reasoning tokens are billed as output tokens.

How is Gemini 3.1 Pro different from Gemini 3 Flash?

Gemini 3 Flash is the fast model of the generation for everyday tasks; Gemini 3.1 Pro is the senior model for the hardest ones. Flash costs 40 / 240 ₽ per 1M tokens, four times less, so pick Pro when you need maximum depth of analysis.

Can I turn reasoning off?

No. Gemini 3.1 Pro accepts only three reasoning effort values: low, medium and high, with low as the default. There is no none value, so the model always thinks a little before answering. On simple questions that is a few dozen tokens: in our test at medium effort it used about 60 reasoning tokens.

Does the model have internet access?

On Genosai, Gemini 3.1 Pro has no built-in web search. The model does not know the latest news, prices or exchange rates, so pass fresh data in the request: paste text or a table, or attach a screenshot.

How do I use Gemini 3.1 Pro through the API?

The Genosai API follows the OpenAI format. Send a POST to https://api.genosai.io/v1/chat/completions with the model gemini-3.1-pro and an sdk_... key. Function calling and the reasoning_effort parameter are supported; the model list is at GET /v1/models.

Try Gemini 3.1 Pro on Genosai