Last night, Gemini launched and open sourced GLM-5, a new DeepSeek model with a larger grayscale test context window (suspected to be an upgraded version of v4). Now it seems that Google may also launch its own new model Gemini 3.1 Pro.
On the social media website X/Twitter, a user posted a comparison screenshot page of the model arena Artificial Analysis. The bottom of the page briefly displayed a reference to the Gemini 3.1 series model in the model comparison.
The name of the referenced model is the Gemini 3.1 Pro Preview model. Since it has a preview label, it means that Google is already testing it. Usually, artificial intelligence developers will open the pre-release version to some core partners or third-party evaluation institutions before the model is officially released after completing training. This way, improvements can be made based on feedback and test data can be obtained before being announced.
However, Gemini is not an open source model. It is estimated that Google will not follow the example of Zhipu and release GLM-5 to OpenRouter under the name Pony Alpha in advance for users to call for free (a few days before the release of GLM-5, Zhipu provided the model to OpenRouter and put it on the shelves under the name Pony Alpha).
The variants offered by Google in Gemini 3.0 include Gemini 3.0 Pro (focused on complex tasks), Gemini 3.0 Flash (offers faster speeds) and Gemini 3.0 Deep Think (deep thinking version), so Gemini 3.1 may also offer the same variants.
At this stage, there is no definite news on when the Gemini 3.1 Pro model will be released. Apart from this quoted screenshot, there is no other data to show the characteristics of the model, so in the end we have to wait for Google to release the model and see the benchmark test results.