New API 是新一代 AI 网关与资产管理系统,作为 AI 基座平台,提供统一基础设施接入全球 30+ 主流 AI 服务(OpenAI、Claude、Gemini、DeepSeek 等)。平台核心特性包括统一 OpenAI 兼容接口、智能路由负载均衡、精细计费与权限管控、实时数据看板。平台支持多格式转换、推理力度控制、缓存计费等高级功能。采用 AGPLv3 开源协议,支持 Docker 一键部署,适配个人开发者到企业级多租户场景。
New API的主要功能
统一接口管理:提供兼容 OpenAI 格式的单一 API 端点,无缝接入全球 30+ 主流 AI 服务提供商。
智能路由调度:支持多渠道负载均衡、故障自动切换和加权随机分发,确保服务高可用性。
精细计费系统:支持实现按次数或按量计费、预付费充值、多倍率配置及缓存计费支持。
安全权限管控:提供令牌分组管理、模型访问限制、API 调用审计及多平台授权登录。
格式转换能力:支持 OpenAI、Claude Messages、Google Gemini 等多种 API 格式之间的相互转换。
推理力度控制:支持通过模型名称后缀灵活设置高、中、低不同级别的推理思考强度。
实时数据看板:提供可视化控制台、用量统计分析和成本监控的数据洞察功能。
New API的关键信息和使用要求
项目定位:新一代 AI 网关与资产管理系统,AI 基座平台
开源协议:GNU AGPLv3(可免费使用,SaaS 部署需开源)
兼容基础:基于 One API 开发,完全兼容原数据库
支持语言:简中、繁中、英文、法文、日文
部署方式:Docker / Docker Compose / 宝塔面板
数据库:SQLite(默认)/ MySQL ≥ 5.7.8 / PostgreSQL ≥ 9.6
Docker 镜像:calciumion/new-api:latest
New API的核心优势
统一接入:通过一个兼容 OpenAI 格式的 API 端点,可无缝接入全球 30+ 主流 AI 服务提供商,彻底告别多平台对接的繁琐工作。
智能路由:平台内置多渠道负载均衡、故障自动切换和加权随机分发机制,确保 AI 服务的高可用性和请求稳定性。
成本优化:支持缓存计费、按量或按次计费以及多倍率灵活配置,帮助用户实现精细化的成本控制和费用管理。
格式互通:提供 OpenAI、Claude Messages、Google Gemini 等多种 API 格式之间的自由转换能力,显著降低不同模型的接入门槛。
开箱即用:支持 Docker 一键部署,完全兼容 One API 数据库,提供宝塔面板可视化安装,极大简化部署流程。
• Claude Code:目前的”天花板”,也是最傲娇的。 Anthropic 对国内用户的敌意真的是写在脸上的。最近 Claude Code 甚至要搞 KYC 认证,只认护照、驾照原件,复印件和照片统统拒收。圈里有个梗:说是不知道当年百度对 Anthropic 的 CEO 做了什么,让人家记恨至今。 梗归梗,但它成本高、风控严是事实。即便如此,大家还是趋之若鹜,这真不是崇洋媚外,只有真正干活的人才知道,顶级模型那种”指哪打哪”的逻辑感,国内模型暂时还听不懂这种”人话”。
The cost of “raising AI crayfish” varies widely, ranging from a few yuan to over a thousand yuan per month, depending primarily on your deployment method and usage intensity.
The term “raising AI crayfish” actually refers to deploying and using an open-source AI agent framework called OpenClaw. The framework itself is free, but running it requires hardware and a “brain” (large language model), both of which incur costs.
💰 Cost Breakdown
The cost of “raising lobsters” is mainly divided into two parts:
Hardware/Server Costs This is the “body” that runs OpenClaw. You have two options:
On-premises deployment: Run it on your own computer. The cost is 0 yuan (using an existing computer), but the computer must remain on 24 hours a day, adding approximately 10–30 yuan to your monthly electricity bill. If you want a more stable experience, you can purchase dedicated hardware (such as a Mac mini), with a one-time investment of about 4,000–8,000 yuan.
Cloud Deployment: Rent a cloud server. This is the more mainstream and simpler approach, with monthly rental costs of approximately 25–150 yuan. Some cloud providers offer promotional deals; for example, Alibaba Cloud previously offered a one-click deployment plan for new users at 9.9 yuan for the first month.
Large Model API Fees (Core Expense) This is the primary ongoing expense for “raising lobsters,” which can be thought of as the “feed” for the lobsters. OpenClaw needs to call upon large language models such as Tongyi Qianwen and DeepSeek to think and execute tasks; these services are typically billed based on usage (tokens).
Light usage (a few simple queries per day): Approximately 70–210 yuan per month.
Regular use (automated office tasks, multitasking): Approximately 210–490 RMB per month.
Heavy use (complex tasks, batch processing): Easily exceeds 700 RMB per month; there are even cases where users spent 1,200 RMB in just 6 hours.
🤔 Is there a more cost-effective way?
Yes. Some vendors have launched pre-packaged solutions that lower the barrier to entry and reduce costs.
Tencent WorkBuddy: Nicknamed the “Tencent version of Xiaolongxia,” it currently offers a free trial version that can be used immediately after registration, without complex deployment or additional API fees. It also offers a paid Professional Edition (58 yuan/month) and other plans.
Alibaba Cloud Plans: Through its one-click deployment solution, combined with free large-model quotas for new users (e.g., over 70 million tokens within 90 days), you can keep costs extremely low in the early stages.
📊 Cost Options at a Glance
For a clearer overview, please refer to the table below: Option Type Hardware/Server Cost Large Model API Fees (Monthly) Features On-premises deployment 0 RMB (Using an old computer) 70–1,000+ RMB Low barrier to entry, but high power consumption and risks Cloud Deployment 25–150 RMB/month 70–1,000+ RMB Stable and secure; the mainstream choice Vendor-Packaged Version 0 RMB (e.g., Tencent WorkBuddy) Starting at 0 RMB (Free version available) Ready to use out of the box, ideal for beginners to try out
⚠️ Special Note: Beware of scams
Recently, there have been many scams under the guise of “AI Lobster Farming.” Please be vigilant:
“Passive Income” Scams: Any “cloud lobster farming” project claiming “passive income without effort,” “daily dividends,” or “high-price buybacks” is an illegal Ponzi scheme.
High-Priced Installation Services: OpenClaw is free and open-source, and there are plenty of free tutorials available online. Anyone charging hundreds or even thousands of yuan for “exclusive installation services” or “paid tutorials” is likely exploiting information asymmetry—or may even remotely control your computer to steal your data.
In summary, for ordinary users handling only routine, light-duty tasks, existing free AI tools are sufficient. Whether you should “invest in a high-end solution” depends on whether you have high-frequency, complex, and automatable tasks that require it, and whether you are willing to bear the corresponding costs.