引言
Google 通过 AI Studio 开放 Gemini 系列模型的免费 API 层，附带每分钟请求数与每日请求上限。免费层支持文本、视觉与多模态输入，是个人项目与学习的首选入口。2026-08-22 起，Gemini 渠道正式加入 APIShare 免费渠道池：10 个对话模型实测入库，全部进入主榜参与评分。
模型与额度
端点：https://generativelanguage.googleapis.com/v1beta/models/{model}:generateContent。渠道池当前入库的 10 个模型：
| 模型 |
输入上限 |
最大输出 |
定位 |
gemini-3.7-flash |
1M |
64K |
当前主力，复杂任务首选 |
gemini-3.6-flash |
1M |
64K |
官方对新项目的推荐稳定款 |
gemini-3.5-flash |
1M |
64K |
稳定代次 |
gemini-3-flash-preview |
1M |
64K |
预览代次尝鲜 |
gemini-3.1-flash-lite |
1M |
64K |
轻量快速，低成本流水线 |
gemini-3.1-flash-lite-preview |
1M |
64K |
轻量快速（preview） |
gemini-flash-latest |
1M |
64K |
自动跟随最新 Flash 的别名 |
gemini-flash-lite-latest |
1M |
64K |
自动跟随最新 Lite 的别名 |
gemma-4-31b-it |
256K |
— |
开源系大杯，通用文本 |
gemma-4-26b-a4b-it |
256K |
— |
开源系轻量款 |
Gemini Flash 系统一 1M 输入 / 64K 输出；gemma-4 双子为 256K 上下文。免费层速率约 15 RPM / 1500 RPD（以控制台为准），突发触发 429 时加令牌桶退避。
ℹ️ 图像生成、TTS、实时语音、embedding、视频/音乐生成等专项能力模型在免费层配额为 0，均未入池——需要专项能力请走付费层，或关注后续入库通知。
调用示例
from google import genai
client = genai.Client(api_key="AIza...")
resp = client.models.generate_content(
model="gemini-3.6-flash",
contents="解释 transformer 的注意力机制",
)
print(resp.text)
注意事项
免费层数据可能用于模型改进（付费层不会）；版本与上下文窗口可能滞后付费层；区域可用性需查文档。生产走付费层。1M 上下文适合长文档摘要与整库代码分析，但 RPD 限制需节制调用频次。切勿使用 gemini-2.5-* 系列（对新用户 API Key 已 404 下线），更早的 gemini-1.5-*、gemini-2.0-flash 也已退出免费层清单，历史代码请迁移至 gemini-3.5-flash 及以上。
Introduction
Google exposes the Gemini family through AI Studio with a free API tier that includes per-minute and daily rate limits. The free tier supports text, vision, and multimodal inputs, making it the go-to entry point for personal projects and learning. As of 2026-08-22, Gemini officially joined the APIShare free channel pool: 10 chat models verified, all entering the main ranking.
Models and Quotas
Endpoint: https://generativelanguage.googleapis.com/v1beta/models/{model}:generateContent. The 10 models currently in the pool:
| Model |
Input cap |
Max output |
Positioning |
gemini-3.7-flash |
1M |
64K |
Current flagship, complex tasks |
gemini-3.6-flash |
1M |
64K |
Google's recommended stable pick for new projects |
gemini-3.5-flash |
1M |
64K |
Stable generation |
gemini-3-flash-preview |
1M |
64K |
Preview generation |
gemini-3.1-flash-lite |
1M |
64K |
Lightweight, low-cost pipelines |
gemini-3.1-flash-lite-preview |
1M |
64K |
Lightweight (preview) |
gemini-flash-latest |
1M |
64K |
Alias that tracks the newest Flash |
gemini-flash-lite-latest |
1M |
64K |
Alias that tracks the newest Lite |
gemma-4-31b-it |
256K |
— |
Open-source workhorse |
gemma-4-26b-a4b-it |
256K |
— |
Open-source lightweight |
The Gemini Flash family ships a uniform 1M input / 64K output envelope; the gemma-4 twins run a 256K context. Free-tier throughput is roughly 15 RPM / 1500 RPD (check the console); back off with a token bucket on 429s.
ℹ️ Specialist capabilities (image generation, TTS, realtime voice, embeddings, video/music) carry zero free-tier quota and are not in the pool — use the paid tier or watch for future intake notices.
Call Example
from google import genai
client = genai.Client(api_key="AIza...")
resp = client.models.generate_content(
model="gemini-3.6-flash",
contents="Explain the attention mechanism in transformers.",
)
print(resp.text)
Caveats
Three constraints matter in practice. First, free-tier request data may be used for model improvement — the paid tier opts out, so sensitive workloads must upgrade. Second, model versions on the free tier can lag the paid tier by a release cycle. Third, regional availability varies — confirm in the docs before building region-specific logic. The 1M context window is excellent for long-document Q&A and repo-scale analysis, but the 1500 RPD cap means pacing batch jobs carefully. Handle prompt_feedback.block_reason explicitly: treating a safety-blocked response as an empty success silently corrupts downstream logic. Finally, never use the gemini-2.5-* family (they return 404 for new users' API keys); older gemini-1.5-* and gemini-2.0-flash have also left the free-tier list — migrate existing code to gemini-3.5-flash or newer.