Cloudflare Workers AI 免费层完全指南：在边缘运行 Llama 3、Mistral 等开源模型

什么是 Cloudflare Workers AI

Cloudflare Workers AI 是 Cloudflare 推出的无服务器 GPU 推理平台，依托其遍布全球 300 多个城市的边缘网络，让开发者可以在距离用户最近的节点运行开源大语言模型、图像生成模型和文本嵌入模型。

免费额度详情

Workers AI 提供永久免费的神经元额度：每日 10,000 Neurons，无信用卡、无时间限制。

资源	免费额度
Neurons	每日 10,000 个
模型种类	50+ 开源模型
速率限制	每分钟 50,000 Neurons

支持的模型

模型	参数量	Neurons/请求
@cf/meta/llama-3-8b-instruct	8B	~134
@cf/meta/llama-3-70b-instruct	70B	~880
@cf/mistral/mistral-7b-instruct-v0.1	7B	~117
@cf/google/gemma-7b-it-lora	7B	~117

快速开始

curl -X POST "https://api.cloudflare.com/client/v4/accounts/ACCOUNT_ID/ai/run/@cf/meta/llama-3-8b-instruct" \
  -H "Authorization: Bearer API_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"messages": [{"role": "user", "content": "Hello"}]}'

性能对比

平台	免费额度	延迟
Cloudflare Workers AI	10K Neurons/天	~300ms
Groq	有限试用	~200ms
Together AI	$5 信用额	~350ms

总结

Cloudflare Workers AI 是最适合个人开发者的免费 AI 推理服务，全球边缘网络、永久免费额度、零门槛接入。

What is Cloudflare Workers AI

Cloudflare Workers AI is a serverless GPU inference platform leveraging Cloudflare's global edge network of 300+ cities.

Free Tier Details

Workers AI offers a permanently free Neurons quota: 10,000 Neurons daily, no credit card, no time limit.

Resource	Free Tier
Neurons	10,000/day
Models	50+ open-source
Rate Limit	50,000 Neurons/min

Supported Models

Model	Parameters	Neurons/Request
@cf/meta/llama-3-8b-instruct	8B	~134
@cf/meta/llama-3-70b-instruct	70B	~880
@cf/mistral/mistral-7b-instruct-v0.1	7B	~117
@cf/google/gemma-7b-it-lora	7B	~117

Quick Start

curl -X POST "https://api.cloudflare.com/client/v4/accounts/ACCOUNT_ID/ai/run/@cf/meta/llama-3-8b-instruct" \
  -H "Authorization: Bearer API_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"messages": [{"role": "user", "content": "Hello"}]}'

Performance Comparison

Platform	Free Tier	Latency
Cloudflare Workers AI	10K Neurons/day	~300ms
Groq	Limited trial	~200ms
Together AI	$5 credit	~350ms

Conclusion

Cloudflare Workers AI is the most suitable free AI inference service for individual developers.