Definition
Gemini Pro is the Pro tier within Google DeepMind's Gemini model family. It includes versioned models used for broad language and multimodal tasks, including complex reasoning and software work. The Pro line spans releases with different architectures and capabilities, so comparisons need a named version and task.
A complete name such as Gemini 3.1 Pro identifies a generation and variant. The API identifier can add preview status or a tool-specific configuration. Preserve those distinctions when comparing results.
Origin and attribution
Google introduced Gemini Pro alongside Ultra and Nano on December 6, 2023. The announcement described Pro as the original family's model for scaling across a range of tasks and credited the Google DeepMind team. Later Pro releases retained the product-line name while changing capabilities.
Scope and limitations
The current Gemini 3.1 Pro documentation identifies a preview model with text, image, video, audio, and PDF inputs and text output. It lists thinking, function calling, and structured output support. The same endpoint lists image generation, audio generation, and the Live API as unsupported. Its capabilities and access terms must be read for that release.
The separate custom-tools endpoint is tuned to prioritize tools such as bash and user-defined functions. Google warns that it can show quality fluctuations on tasks that do not benefit from those tools. A configuration suited to one agent harness therefore needs evaluation in another.
Operational significance
Choose Pro using measured task quality and total cost, including reasoning and tool calls. Compare it with an appropriate Flash or Flash-Lite release on the same inputs and acceptance criteria. A newer generation, different tool interface, or different reasoning setting can change the comparison.
Keep permission and verification checks in the application. A model's supported tool-call format allows it to propose an invocation; the factory decides whether to execute that invocation and whether its result is sufficient.
Distinguish it from nearby terms
- Gemini is the whole model family and also an application name; Pro is one tier within that family.
- Flash and Flash-Lite are separate product lines with their own operating goals and version histories.
- A reasoning model describes inference behavior. Pro identifies a vendor tier whose specific reasoning interface depends on the release.
Check your understanding
A team routes every difficult request to Pro. Which workload evidence would justify that rule, and what changes if the selected endpoint is a preview release tuned for custom tools?