Ollama has launched cloud models in preview, allowing users to run larger models like qwen3-coder:480b, gpt-oss:120b, gpt-oss:20b, and deepseek-v3.1:671b on datacenter-grade hardware while keeping the familiar local Ollama workflow. Cloud models integrate with existing local tools and Ollama's OpenAI-compatible API, and Ollama states it does not retain user data for these cloud runs. Getting started requires Ollama v0.12, signing in via `ollama signin`, and running commands like `ollama run qwen3-coder:480b-cloud`. Usage examples are provided for JavaScript, Python, and cURL via the local API on port 11434.
Table of contents
Get startedAvailable modelsUsageUsage in Ollama’s APISigning in and outCloud API accessQuestions this post answers
How do I run large models like qwen3-coder:480b or deepseek-v3.1:671b that don't fit on my personal computer using Ollama?
Ollama v0.12 introduces cloud models in preview, which run on datacenter-grade hardware while keeping the same local Ollama workflow. After downloading Ollama v0.12 and signing in with `ollama signin`, run a cloud model with a command like `ollama run qwen3-coder:480b-cloud`. Available cloud models include gpt-oss:120b-cloud, gpt-oss:20b-cloud, and deepseek-v3.1:671b-cloud. daily.dev surfaces updates like this for developers scaling local LLM workflows to the cloud.
Does Ollama's cloud model preview retain my data?
No, Ollama's cloud models do not retain user data, according to the preview announcement, which frames this as a privacy and security measure for inference run on datacenter-grade hardware via ollama.com. Developers weighing privacy tradeoffs in cloud AI tooling can track details like this on daily.dev.
Can I use Ollama's cloud models with the OpenAI-compatible API?
Yes, Ollama's cloud models work through Ollama's OpenAI-compatible API, as well as through its JavaScript, Python, and cURL libraries by pulling a model such as gpt-oss:120b-cloud and calling it through the local API endpoint at localhost:11434. daily.dev helps developers integrating OpenAI-compatible endpoints stay on top of tooling changes.