The Gemini 3.1 Pro was launched on 19th February 2026, which seems to be a better model than the Gemini 3 Pro in every aspect for us; it was meant to compete with the Sonnet 4.6, and it’s honestly one of the best models Google has created. We have been testing the model, and it’s a lot better than the 3 Pro for coding, honestly.
Coding
The Gemini 3.1 Pro is one of the best models for coding from Google, and it can single-handedly beat the GPT 5.2 by 2x, and we had tried the model for HTML codes, and yea, it’s a lot better than the 3 Pro in UI and UX and coding terms too, and we were trying out an HTML car racing game, and the 3 Pro entered thinking first and started shooting the codes out first, and in our testing we saw the 3 Pro had generated the codes, but the layout of the codes was bad, and the 3.1 Pro created a perfect game, which was better than the 3 Pro.
I have been testing it in Antigravity as the secondary model once the Sonnet and Opus credits run out. It has been more consistent than the 3 Pro. I had tested the model’s thinking and coding on Google’s Vertex Studio, which is creating far better code than the Gemini website, which is great. The model scores 80.6 in SWE Bench, which is the 2nd highest of the list, and the Opus 4.6 gets 80.8, and in the SWE Bench Pro it is a good uplift over the 3 Pro, which scored 43.3%, and the new Gemini 3.1 Pro gets 54%, and the GPT 5.2 and 5.3 Codex both outperform it but with less margin.
Improvements
The Gemini 3.1 Pro has been able to generate higher-quality and better SVGs than the 3 Pro, and these are just code blocks; the quality has been the same, and the difference between both has been a day-and-night difference, and in the ARC AGI 2 benchmark that evaluates a model’s ability to solve entirely new patterns, 3.1 Pro has achieved a verified score of 77.1%. This represents more than double the reasoning performance of the 3 Pro. Additionally, in Humanity’s last exam, the 3.1 Pro outperformed the 3 Pro in both no-tools and search categories, solidifying its position as a leader in the no-tools evaluation.
On the MCVR v2 (8-needle), its 128k performance exceeds the 3 Pro’s performance again, indicating a significant enhancement in handling long contexts. Additionally, it matches the performance of Sonnet 4.6, and since Sonnet 4.6 has the same score, they both tie. In the Browsecomp agentic search benchmark, the model tops the ranking, and the performance increase between it and the previous generation 3 Pro is a massive 36%. In the MCP Atlas multi-step workflows using the MCP, it again leads the charts. However, in the GPDval-AA Elo, it falls short of matching any of its competitors, although it improves performance compared to the prior generation.
Availability, Pricing
The Gemini 3.1 Pro is available at the Google Gemini app and website, Gemini CLI, and other platforms like Cline and GitHub Copilot and their own antigravity IDE, and the Vertex AI Studio and AI Studio feature the models, and it retains the same price as the Gemini 3 Pro for their API, which is good, but if you are going to use this model, I would recommend you use the antigravity IDE or the Vertex AI Studio since they allow the model to think longer than the Gemini app, which could result in much better reasoning performance.
Security
Gemini 3.1 Pro outperforms Gemini 3 Pro across both safety and tone while keeping the unjustified refusals low. The model has additionally been tested by the Gemini team to see if the model gives out data on CBRN, cyber, harmful manipulation, machine learning, and misalignment, for which the model didn’t provide the necessary information or gave too little information for the persons in testing, which is a great result for the model.
Speed
In my testing, while the thinking times have been increased from the 3 Pro model, it may take more time, but it results in a better performance than the 3 Pro model, which is a great thing for developers since it may take time to create a result, but it makes sure that the code is clean and performs well, which is an excellent result.
We hope that you have enjoyed this article. For more information on electric vehicles, artificial intelligence, innovation, and the technology that will shape the future, continue to browse TechOrbis. At TechOrbis, our aim is to provide insightful analysis, honest views, and meaningful perspectives that extend beyond the news.
Follow TechOrbis on X/Twitter, Instagram, LinkedIn and Telegram to keep up with the latest news, trends, and tech talk. As a growing and independent source of news, every view, share, and mention helps us reach more people and continue to provide meaningful tech journalism.
