People are still talking about the funny scene where the bosses of OpenAI and Anthropic refused to hold hands and even clenched their fists in the air. On the other hand, Google iterated on the model with a backhand.


And this iteration is quite like a wolf in sheep's clothing - judging from the numerical serial number of the model suffix, this is the "smallest" iteration of Google. Previous iterations were from Gemini 2.0 to Gemini 2.5, this time from Gemini 3.0 to Gemini 3.1Pro Preview.


But this iteration of ".1" has not improved at all.

Google CEO Sundar Pichai said that the new generation model is very good at handling "super complex tasks." Such as visualizing complex concepts, synthesizing data into a single view, or bringing creative projects to life.


Yao Shunyu also specially posted on X to cheer up the Gemini 3.1 Pro Preview and praised:

"Gemini is not just a good model, better models are coming with an unstoppable trend."


It should be noted that about a week ago, Google launched the "dedicated reasoning mode" Gemini 3 Deep Think, which is designed for complex, open-ended problems such as science, research, and engineering.

Demini 3 Deep Think is the first project Yao Shunyu participated in after switching from Anthropic to Google DeepMind.

Today’s Gemini 3.1 Pro Preview is inextricably related to Gemini 3 Deep Think. The official stated that it is “directly built on the experience and technology of Gemini 3 Deep Think.” It is equivalent to "delegating" Deep Think's core reasoning improvement technology to the more widely available Pro model.

01

What Gemini 3.1 Pro Preview can do

Since the outstanding ability of this new generation model is to handle "super complex" tasks, ordinary conversations are put aside. Google's official blog post focuses on several examples to show off its muscles.

First, create SVG animations through simple prompt words.

This function was also available in the previous generation, but it has improved significantly in comparison.

For example, the prompt "Generate an SVG depicting a chameleon sitting quietly on a branch. Let the chameleon's eyes follow the user's mouse cursor as it moves on the screen."

The animated background generated by Gemini 3 Pro is a single white color, and the chameleon also looks dull, even with two eyes on one side.

The animation generated by Gemini 3.1 Pro has a rich "dark green jungle" background. The chameleon's body is decorated with yellow markings and dots, its eyes are three-dimensional, and its legs are naturally curved.


Another example is the prompt word "Generate an SVG of a sliding toggle switch that turns the sun icon into a glowing moon when the mouse is hovered over it, while the background smoothly gradients from bright to dark. Adopt a clean flat UI style."

Although the animation given by Gemini 3 Pro has completed the task, and the icon can change with the mouse, the main icon is single, a circular pattern with missing corners, with yellow representing daytime and white representing nighttime.

The animation generated by Gemini 3.1 Pro is much more complex. During the day, it is a yellow sun and white clouds, and at night, it is a crescent moon and stars. The two sets of icons change smoothly.


All in all, the animations produced by Gemini 3 Pro are reminiscent of the “I studied animation for three years” meme many years ago.


The SVG animation delivered by Gemini 3.1 Pro has reached the point where it can be used directly.

Second, build an engineering-level system.

Gemini 3.1 Pro can already directly generate a complete interactive system integrating 3D rendering, real-time solar ephemeris calculation, API asynchronous pulling and physical light effects based on a highly complex natural language command, instead of a simple page demo.

In the example given by Google, the user gives text instructions, and Gemini 3.1 Pro generates and builds a high-fidelity, interactive 3D International Space Station (ISS) orbit tracker. Render a detailed 3D Earth model using high-resolution Blue Marble texture maps.


Third, generate interactive creative systems.

In another example, Google showed off a complex 3D starling murmuration simulation written in Gemini 3.1 Pro.


It not only generates visual code, but also builds an immersive experience where users can control the birds through hand tracking and listen to a generative soundtrack that changes according to the movement of the birds.

For researchers and designers, this provides a powerful way to prototype sensory-rich interfaces.

Fourth, convert literary themes into runnable code.

This example may be the easiest one for ordinary people to get the power of.


When asked to create a modern personal portfolio website for Emily Brontë's Wuthering Heights, the model did not simply summarize the text content, but reasoned based on the atmosphere and mood of the novel, designed a simple, contemporary interface, and created a website that captures the protagonist's spiritual core.

There is no need to say much about the value of abstract reasoning.

02

How strong is it?

New generation models inevitably go through the step of ranking.

The upgrade of ".1" has achieved multiple jumps at every turn.

According to the test results released by Google’s official blog post,

In the ARC-AGI-2 benchmark, the 3.1 Pro achieved a verification score of 77.1%. The inference performance is more than doubled compared to 3 Pro.

This is also consistent with the 3.1 Pro example, as this test evaluates the model's ability to solve completely new logical patterns. In human terms, it is the ability to abstractly reason and solve puzzles.


In addition, in the GPQA Diamond (scientific knowledge test), 3.1 Pro scored 94.3%; on the intelligent agent benchmark MCP Atlas, it scored 69.2%; on BrowseComp, the benchmark for real network browsing and information integration capabilities, it scored 85.9%.

These results exceed Anthropic's Sonnet 4.6, Opus 4.6, and OpenAI's GPT-5.2 and GPT-5.3-Codex.

GoogleGemini 3.1 Pro has significantly widened the gap in ARC abstract reasoning and BrowseComp search tasks this time, showing a clear tendency to be an agent rather than a pure knowledge model.

In addition, a third-party evaluation agency specializing in large model benchmark testing and comparative analysis also released relevant test results, praising Gemini 3.1 Pro for leading 6 of the 10 evaluations that make up the Artificial Analysis Intelligence Index. Compared with Gemini 3 Pro Preview, it has significantly improved in many capabilities, especially in reasoning and knowledge, coding capabilities, and reducing hallucinations.


Moreover, Gemini 3.1 Pro Preview maintains high token efficiency.

Running the full Artificial Analysis Intelligence Index requires approximately 57 million tokens (1 million more than Gemini 3 Pro Preview).

This token usage is lower than other cutting-edge models running in maximum inference mode, such as Opus 4.6 (max) and GPT-5.2 (xhigh).

Combined with lower per-token pricing, Gemini 3.1 Pro Preview has a cost advantage among leading-edge models, with the cost of running a full Intelligence Index being less than half that of Opus 4.6 (max), but still approximately twice that of the leading open source model GLM-5.

03

Capacity doubled, price unchanged

Google’s official API pricing shows that the charging structure of Gemini 3 Pro/3.1 Pro Preview is based on tokens:

For less than 200k tokens, the input price is about $2 per million tokens, and the output price is $4. When there are more than 200k tokens, the input is US$4 per million tokens and the output is US$18.

For context caching, there is a charge of $0.20 to $0.40 per million tokens, depending on the size of the prompt word, plus an hourly storage fee of $4.50 per million tokens.

This price is generally consistent with Gemini's own previous generation 3 Pro, but compared with the Anthropic Opus series, it is still relatively cheap. The input/output unit price of models like Opus can be around $5/$25.

Especially combined with its current outstanding model capabilities, the price is even more competitive.

Don’t forget, what Google released this time is only a “Preview”, and Google will launch the official version soon. As for the ".1" iteration, Google is also hinting that it is just showing off its muscles.

Currently, developers can use 3.1 Pro on AI Studio, Gemini API, Gemini CLI, intelligent agent development platform Google Antigravity, and Android Studio; enterprise users can use it on Vertex AI and Gemini Enterprise; ordinary users can use it on Gemini applications and NotebookLM, but the latter is limited to Pro and Ultra subscribers.

There are already many people in various communities who can't wait to get started. It's really just like Google's demonstration, and they have a lot of amazing things.

Someone used Gemini 3.1 Pro to generate an interactive 3D mechanical-level automobile suspension system simulator, which includes real geometric structure, link constraints, and real-time steering and stroke calculations. It is equivalent to writing mechanical engineering modeling, physical logic, and 3D visualization into a runnable tool at once, which is close to engineering-level prototype capabilities.


Someone used 3.1 Pro to create a loop animation of "Ghost Hunter Walking Through a Haunted House" and exclaimed "Gemini wasn't kidding."


In short, Google really held back a big move this time.

A small ".1", but its reasoning and coding capabilities are soaring, and the pricing is so stable.

The overflowing enthusiasm in the community for hand-rubbing demos also proves its capabilities and practicality.

The AI ​​circle is becoming more and more realistic. No matter how powerful the model is, it all depends on whether the bill is worth it. Enterprises begin to carefully calculate the return of each token, and developers also have to weigh the cost-effectiveness. This step by Google is not only to regain the throne, but also to push the competition to a new stage of "who can live better?"

Let’s see how Anthropic and OpenAI, as well as xAI, Meta, Microsoft and other competitors who are clenching their fists, will respond.