Gemini 3 Pro Preview
Gemini 3 Pro Preview is the flagship reasoning model in the Gemini 3 generation for demanding agentic and analytical tasks, with improvements in multi-step function calling, complex image reasoning, long-document analysis, and instruction following over Gemini 2.5 Pro.
import { streamText } from 'ai'
const result = streamText({ model: 'google/gemini-3-pro-preview', prompt: 'Why is the sky blue?'})What To Consider When Choosing a Provider
Zero Data Retention
AI Gateway supports Zero Data Retention for this model via direct gateway requests (BYOK is not included). To configure this, check the documentation.Authentication
AI Gateway authenticates requests using an API key or OIDC token. You do not need to manage provider credentials directly.
Gemini 3 Pro Preview is a reasoning model: enable includeThoughts via providerOptions.google.thinkingConfig to surface the model's reasoning trace, which is particularly useful when auditing complex multi-step outputs.
When to Use Gemini 3 Pro Preview
Best For
Multi-step agentic workflows:
Sequential function calls that require reliable planning and execution
Deep document analysis:
Combining long text with embedded charts, diagrams, and images
Instruction-following tasks:
Precision and completeness are critical to downstream correctness
Reasoning-intensive applications:
Surfacing the model's thought process aids auditability
Complex technical research:
Tasks requiring synthesis across disparate sources and formats
Consider Alternatives When
Latency and cost primary:
Per-token cost and response speed dominate (consider
google/gemini-3-flashfor pro-grade quality at flash speed)Updated agentic quality needed:
For the latest improvements on software engineering tasks (consider
google/gemini-3.1-pro-preview)Native image generation output:
Your workflow requires image output (consider
google/gemini-3-pro-image)High-volume straightforward tasks:
Extraction or translation at scale (consider
google/gemini-3.1-flash-lite-preview)
Conclusion
Gemini 3 Pro Preview targets tasks where getting every step right matters more than getting the answer quickly. That means complex agentic pipelines, technical document analysis, and multimodal reasoning that spans images and long text. For teams building the highest-stakes AI features, this is the Gemini 3 model designed for reasoning depth rather than maximum throughput.
FAQ
Four specific improvements: multi-step function calling, planning, reasoning over complex images and long documents, and instruction following. These directly address the reliability gaps that affect agentic workflows at scale.
Set includeThoughts to true under providerOptions.google.thinkingConfig in the AI SDK. Use streamText for streaming, and the model emits reasoning tokens alongside the generated response.
It can be, but it is a reasoning model with higher latency than the Flash tier. For interactive applications where sub-second responses are required, google/gemini-3-flash provides pro-grade reasoning at significantly lower latency.
Yes. The model handles long documents with embedded charts, diagrams, and images. Improved reasoning over complex images and long documents is one of its headline capabilities over Gemini 2.5 Pro.
Gemini 3.1 Pro introduces additional quality improvements for software engineering and agentic tasks, enhanced usability for finance and spreadsheet applications, and more efficient thinking that reduces token consumption. Gemini 3 Pro Preview was the initial release; 3.1 Pro builds on that foundation.
No. AI Gateway manages all underlying provider credentials. You authenticate once using a Vercel API key or OIDC token.
The model more reliably executes sequences of tool calls: choosing the right tool, interpreting its output, deciding whether to call another tool, and knowing when the task is complete. This reduces the need for human intervention to correct routing errors mid-workflow.
Yes. You can pass image inputs alongside text prompts to enable cross-modal analysis within a single request.
