Gemini 3 Pro
gemini-3-pro-openai

Gemini 3 Pro API 是 Google DeepMind 推出的新一代多模态模型,具备卓越的推理能力,能无缝理解文本、图像、视频及音频,并支持大规模长上下文处理。

Pricing: Input 100 credits / 1M tokens (≈ $0.50), Output 700 credits / 1M tokens (≈ $3.50). Both are ~70–75% cheaper than official pricing. High-tier top-ups (+10% bonus) bring effective pricing down to ~$0.45 for Input 1M tokens and ~$3.15 for Output 1M tokens.
gemini-3-pro-openai

Automatically append previous messages to maintain multi-turn context. May increase token usage.

Structured outputs
Function calling
Grounding with Google Search
Source:Google Search
Thinking process
Reasoning effort

How can I help you today?

README

Complete guide to using gemini-3-pro

高性价比 Gemini 3 API 与 Gemini 3 Pro API 接入方案

Original image
Google Gemini 3 Pro API:具备业界顶尖的推理能力
通过 Gemini 3.0 API 深入理解文本、图像、文件与视频
借助 Gemini 3 API 打造高保真可视化效果与丰富交互体验
基于 Gemini 3 API 的进阶智能体编程与工具执行
功能维度评估指标 / 基准Gemini 3 Pro 得分开发者核心洞察
Academic ReasoningHumanity’s Last Exam 37.5%Demonstrates PhD-level reasoning and reliability on complex problems.
Scientific KnowledgeGPQA Diamond (No Tools)91.9%Industry-leading factual accuracy for technical and research tasks.
Multimodal FusionMMMU-Pro (Multimodal Reasoning)81.0%World-leading ability to analyze and synthesize diverse inputs (image + text).
Video UnderstandingVideo-MMMU (Knowledge Acquisition)87.6%Exceptional comprehension of real-world video data and temporal context.
Agentic CodingTerminal-Bench 2.0 (Tool Use)54.2%Superior execution and validation for autonomous coding workflows.
Long-Context ReasoningMRCR v2 (1M-token needle test)26.3%Efficient retrieval and reasoning within massive 1-million-token context windows.
Competitive CodingLiveCodeBench Pro (Elo Rating)2,439Highest score for producing accurate, high-quality code under constraints.
Long-Horizon PlanningVending-Bench 2 (Net Worth)$5,478.16Sustained strategic decision-making over multi-step workflows using gemini 3 api.
评估指标 / 基准Gemini 3 Pro 得分GPT-5.1 得分性能提升
Reasoning (Humanity’s Last Exam)37.5%26.5%~41% higher accuracy in complex, PhD-level problem solving.
Agentic Coding (Terminal-Bench 2.0)54.2%47.6%Superior reliability in executing autonomous coding tasks.
Long-Horizon Planning (Vending-Bench 2)$5,478.16$1,473.433.7× higher value generation for extended agentic workflows.
Multimodal Understanding (MMMU-Pro)81.0%76.0%Leading ability to synthesize text, image, and video inputs.
Context Window1M TokensNot Publicly ListedMassive context for processing enterprise-scale datasets.
评估指标 / 基准Gemini 3 Pro 得分Claude Sonnet 4.5性能提升
Context Window1M Tokens200K Tokens5× larger context capacity for analyzing massive documents or codebases.
Reasoning (Humanity’s Last Exam)37.5%13.7%Drastically better performance on difficult, nuance-heavy cognitive tasks.
Multimodal Understanding (MMMU-Pro)81.0%68.0%Superior accuracy in cross-modal (image/text/video) reasoning.
Long-Horizon Planning (Vending-Bench 2)$5,478.16$3,838.74More consistent and profitable decision-making over time.
4.9/ 5
47,068 ratings
Tap a star to rate

Utilize Gemini3 Pro API’s superior agentic capabilities for complex automation. Developers can build autonomous assistants that execute multi-step tasks across external tools, manage large-scale data processing, or handle long-horizon planning. The consistent tool use and deep reasoning make it ideal for dependable enterprise automation.

Accelerate front-end development using Gemini 3 Pro coding capabilities. The model masters Vibe Coding, transforming ambiguous prompts, sketches, or complex instructions into functional, interactive web UIs and code snippets instantly. This capability drastically reduces development cycles and allows designers and developers to collaborate more fluidly.

Go beyond simple text generation. Gemini 3.0 API allows for deep multimodal fusion, enabling applications to analyze video content, interpret image inputs, and generate accompanying code or detailed descriptions. Use the Google's Gemini 3 Pro API to create adaptive learning platforms, comprehensive visual search tools, or automated content summarization engines.

Tap into the model's PhD-level reasoning for high-stakes analysis. The vast 1M context window and high accuracy make the gemini-3-pro API perfect for analyzing extensive research papers, complex legal documents, or large financial reports. Developers can build tools for expert-level knowledge extraction and algorithmic development.

Gemini 3 Pro API 在 Gemini 2.5 Pro 的基础上进行了全面升级,提供了显著增强的推理能力、扩展的多模态能力(涵盖图像、视频、音频及文本),以及超大的上下文窗口。Google 此后发布了 Gemini 4 Carbon,并称其为自家最强的推理模型;Kie.ai 上也可以使用。

您可以在 Kie AI 上使用试用额度免费体验 Gemini 3 Pro API。只需注册账户,获取 API 密钥 (API Key),即可立即开始使用 Gemini API。

基准测试表明,Google 的 Gemini 3 Pro API 在关键领域超越了 GPT-5.1:在“Humanity’s Last Exam”(人类终极考试)测试中得分约为 37.5%(GPT-5.1 为 26.5%),展现出更强的智能体规划能力。此外,在处理复杂的视频与图像分析时,其原生多模态融合能力也获得了显著更高的评价。

接入简单,对开发者十分友好。Kie AI 提供了详尽的开发文档助您快速上手。您只需通过 Kie AI 获取 API 密钥,即可访问 Gemini 3 Pro API,方便您在自己的应用中轻松使用该模型。

当然支持。作为原生多模态模型,Gemini 3 API 不仅支持文本,还能直接接收视频、音频和图像输入。它能够分析视频流、解析音频的细微之处,并将其与文本指令实时关联,为内容平台提供卓越的多媒体理解能力。

Gemini 3 Deep Think 是一种增强型推理模式,在 GPQA Diamond 等复杂基准测试中,其表现优于标准的 Pro 模型。它专为解决需要深度思考过程的极高难度问题而设计,能够为前沿研发应用提供更深层的洞察。

Gemini 3 Pro 是 Gemini 3 API 的标准模型,主打快速响应;而 Gemini 3 Deep Think 则是 Gemini 3 Pro API 下的一种推理模式,通过更深度的思考过程,在处理复杂任务时提供更精准的结果。