🖥️ 本地部署

Ollama 免费吗?价格、免费额度与海外可用性

本地运行开源大模型的事实标准:一条命令拉取 Llama、Qwen、DeepSeek、Gemma 等模型,离线推理、OpenAI 兼容接口、数据不出本机。另提供 Ollama 云端(按 token 计费 + 订阅套餐),本地算力不够时可直接调云端大模型。

The de facto standard for running open models locally: pull Llama, Qwen, DeepSeek or Gemma with one command, run offline with an OpenAI-compatible API and data that never leaves the machine. Ollama Cloud is also available — pay-per-token plus subscription plans — for when local hardware is not enough.

这个工具擅长什么 / Highlights

本地一条命令跑模型,同时能把同一套接口切到云端大模型

One command locally, and the same interface can switch to cloud models

价格与免费额度 / Pricing & free tier

本地永久免费开源;云端按量 $0 起,Pro $20/月、Max $100/月、Team $500/月(含月度额度)

详细介绍 / Overview

本地运行开源大模型的事实标准:一条命令拉取并运行 Llama、Qwen、DeepSeek、Gemma 等主流模型,无需配置环境,Windows / macOS / Linux 全支持;自带 OpenAI 兼容接口,各类客户端可直接调用,模型全部跑在本机、数据不出设备。除本地外还提供 Ollama 云端:同一套接口可切换到托管模型,按 token 计费(DeepSeek、GLM、Kimi、gpt-oss 等在列),并分 Pro $20/月、Max $100/月、Team $500/月 等订阅档,各含月度额度。

站长心得 / Our take

本地跑模型最省事的方案,没有之一:一条命令完成部署,OpenAI 兼容接口让现有工具无缝接入。

好在哪:跨平台一致、模型库更新快、命令行和 API 都能用,团队里不懂底层的同事也能上手。

要注意:它吃显存,7B 以上就要看显卡;默认上下文只有 4096,做长文档问答必须先调 num_ctx,不然会莫名截断;并发默认 1 路,多人共用会排队,得配 OLLAMA_NUM_PARALLEL。

关于云端:本地算力不够时能切到托管模型,按量或订阅都行,省了自己搭 vLLM 的功夫;但云端价格不一定比直接调用厂商 API 便宜,模型选择也受它的目录限制,真跑量还是建议自己比价。总体是“先把本地跑起来”的最优起点,规模化再换 vLLM / SGLang。

Overview (English)

The de facto standard for running open models locally. One command pulls and runs Llama, Qwen, DeepSeek or Gemma on Windows, macOS or Linux, with an OpenAI-compatible API so existing clients work unchanged and data never leaves the machine. Ollama Cloud is also available for when local hardware is not enough — pay-per-token or subscription plans.

Our take (English)

The right first step for local inference, and the honest limits matter.

Where it wins: setup friction is close to zero — one command and you are running, and the OpenAI-compatible API means you can point existing tools at it with a config change. Cross-platform consistency is good and the model library updates quickly.

Where to be careful: three things bite people. It is VRAM-hungry — anything above 7B needs a serious GPU. The default context is only 4096, so long-document work silently truncates until you raise num_ctx. And concurrency defaults to one request, so shared use queues up.

Verdict: the best starting point for local models and private experiments. If you need throughput or multi-user serving, move to vLLM or SGLang — Ollama is a workstation tool, not a server.

本地部署 开源模型 隐私 云端

常见问题 / FAQ

Q:这个工具免费吗?

有免费档位:本地永久免费开源;云端按量 $0 起,Pro $20/月、Max $100/月、Team $500/月(含月度额度)

Q: Is it free?

A: Yes, there is a free tier: 本地永久免费开源;云端按量 $0 起,Pro $20/月、Max $100/月、Team $500/月(含月度额度)

Q:免费额度有多少?

本地永久免费开源;云端按量 $0 起,Pro $20/月、Max $100/月、Team $500/月(含月度额度)

Q: How much free quota is there?

A: 本地永久免费开源;云端按量 $0 起,Pro $20/月、Max $100/月、Team $500/月(含月度额度)

Q:支持海外访问吗?

全球可访问;个别服务可能需要当地账号或网络条件。

Q: Is it available overseas?

A: Globally accessible; some services may need a local account or specific network conditions.

访问 Ollama ↗ ← 返回导航首页