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。

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。

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.

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

Where it winsSetup 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 carefulThree 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.

VerdictThe 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.