Running Qwen3-Coder:480B in VSCodium with Ollama Cloud

For a while I’ve been running Ollama locally together with OpenWebUI.

The setup works great, but my hardware has its limits. My machine uses an RTX 3070, which means I can comfortably run models up to around 8B parameters. Anything larger quickly becomes slow and impractical for everyday use.

While experimenting with Ollama, I discovered that cloud-hosted models can be accessed through the Ollama API:

https://ollama.com/v1

Even on the free tier, the performance and model quality were significantly better than what I could run locally. Suddenly, models that would normally be impossible on my hardware became usable.

That completely changed the game.

Looking for a Coding Agent

My next step was experimenting with coding agents.

I tried Google’s Antigravity, but I quickly ran into token limits. Since I was already using Ollama, I started looking for something that would integrate with my existing setup instead of adding another platform.

Because I work mainly in VSCodium, I explored extensions and eventually found Continue.

One thing I liked immediately was that Continue can connect directly to an Ollama API instance.

My local models worked without any issues.

The cloud models were another story.

The Problem

I wanted to use Qwen3-Coder 480B through Ollama Cloud.

In OpenWebUI the model was available and worked perfectly.

But in VSCodium (via Continue), the model did not work.

My configuration looked like this:

name: Local Config
version: 1.0.0
schema: v1
models:
  - name: qwen3-coder
    provider: ollama
    model: qwen3-coder:480b
    apiBase: http://localhost:11434
    roles:
      - chat
      - edit
      - apply

All my local models worked fine.

But the cloud models didn’t work, even though they were correctly configured in Continue.

At first I assumed it was a configuration issue, but after testing different setups, that wasn’t the case.

The Solution

After some digging, I found the fix.

Ollama requires cloud models to be registered locally before some clients can see them.

The solution was simple:

ollama pull qwen3-coder:480b-cloud

This does NOT download the model locally.

Instead, it registers the cloud model so it becomes visible through the local Ollama API.

After i just setup the model in VScodium this way

name: Local Config
version: 1.0.0
schema: v1
models:
  - name: qwen3-coder-cloud
    provider: ollama
    model: qwen3-coder:480b-cloud
    apiBase: http://localhost:11434
    roles:
      - chat
      - edit
      - apply

After that, the model started working properly.

What I Learned

Even though OpenWebUI can access cloud models directly, other tools using the local Ollama API may not automatically see them.

Key takeaway:

Cloud models may need local registration ollama pull is not always about downloading Some tools depend on the local model registry, not just the API

If a cloud model is missing, try:

ollama pull qwen3-coder:480b-cloud

Final Thoughts

What started as a way to run better models than my hardware allows turned into a more flexible workflow:

OpenWebUI for experimenting with models Ollama for managing both local and cloud models Continue in VSCodium for coding assistance Ollama Cloud for access to large models

This setup feels like a good balance between local control and cloud capability.

This is the beginning of my AI Playground series, where I’ll keep documenting small but useful discoveries while working with AI tools.