Gemini 3.1 Pro model details for developers

Google announced the release of Gemini 3.1 Pro on Feb 19th 2026. The new pro model is already making waves and beats pretty much all other models at the time of release on various benchmarks. Consider taking a look at Gemini 3.1 pro if you are doing some super complex reasoning and want the best of the best. For coding, the 2 most important benchmarks are SWE-Bench and Terminal-Bench. Below you'll find an extracted version of the official benchmarks that zooms in on just those 2 coding benchmarks. Notice that the only LLM to beat Gemini 3.1 Pro as of today is Codex, specifically, GPT-5.3-Codex, with the Extra High reasoning effort enabled.

Interesting facts from the related benchmarks:

• Terminal-Bench leader is GPT-5.3-Codex (xhigh) at 77.3%.
• SWE-Bench Verified is tightly packed: 76.2% to 80.8% (only 4.6 points spread).
• SWE-Bench Pro (Public) leader is GPT-5.3-Codex (xhigh) at 56.8%, with GPT-5.2 close at 55.6%.
• Within GPT-5.2 rows, GPT-5.2 (xhigh) beats GPT-5.2 on Terminal by +8.2 (62.2 vs 54.0).
• Gemini 3.1 vs 3 gains are large: +11.6 (Terminal), +4.4 (SWE Verified), +10.9 (SWE Pro).
• Cross-model ranking is still partial because several rows have missing SWE entries (-).

3.1 Pro is rolling out across a few spots: for developers, it’s in Google AI Studio (via the Gemini API), Gemini CLI, Google Antigravity, and Android Studio (preview). For companies, it’s in Vertex AI and Gemini Enterprise. For regular users, it’s in the Gemini app and NotebookLM.