Ollama Cloud Models Can Be Switched with One Command, but One Message Used 28% of the Quota
Ollama runs frontier open-source models in the cloud, and the commands are the same as local. The tweet @mentioned CrusoeAI and @mchiang0610. Behind the convenience is token-based billing; a user reported that a single Kimi K3 message consumed 28% of the five-hour quota. Look at the bill first, then switch models.
Ollama has moved frontier open-source models to the cloud, and the commands have not changed. The tweet is direct: starting today, you can run the best frontier open-source models on Ollama's cloud, and it @mentioned CrusoeAI and @mchiang0610.
In the video accompanying the tweet, the person on camera is wearing a black jacket, with outdoor hills in the background. The original link is here.
Ollama was previously more often treated as a tool for running open-source models locally. Now the value of the cloud is very concrete: the same API, the same command, and even if the model is too large for local VRAM, you do not need to change computers first. Someone in the replies called this lowering the "laptop tax," giving people who want the Ollama experience but do not want to max out local VRAM a bridge.
The convenience is also here. One netizen said that local and cloud use the same Ollama commands, so it will be easy to pull things back to local later. Another reply mentioned that cloud models and local models go through the same Ollama API; if an app wants to try a large model that a laptop cannot run, you just change the model name.
But the bill changes how it feels to use. A reply reminded people that starting August 31, Ollama's Pro, Max, and Team plans will be billed by token, and each tier includes some usage; local tools remain unchanged, while models run on Ollama's hardware. Right after that, someone asked whether the limits of the old Pro plan had been lowered: they sent only one message to Kimi K3 and it consumed 28% of the five-hour usage quota. This feedback is very specific and puts the cost issue of cloud models front and center.
Another reply calculated task cost: across 100 hacking labs, the best open-source model cost about 24 cents per problem solved, using deepseek 4.1 flash max, with zero refusals, and they hope this model will also come to Ollama. Others care about the billing path: do Ollama cloud models go through your own provider key, or through the Ollama account?
Put together, these replies paint a clear picture. Ollama Cloud compresses the action of "trying a large model" into changing one model name, and the price is that before trying it, you had better look clearly at the token bill. The more frontier the model, the easier it is to make people act quickly; the finer the quotas and billing, the easier it is to make people stop. The cloud threshold is not registration; it is the bill.
发布时间: 2026-10-28 21:48