Integrations
Roo Code
Roo Code in VS Code with AI Tokens: the OpenAI Compatible provider, a model that calls tools, and the context window to set.
Last updated: 2026-10-11
Settings
In Roo Code's settings:
| Field | Value |
|---|---|
| API Provider | OpenAI Compatible |
| Base URL | https://api.aitokens.ch/v1 |
| API Key | your key, sk-at-… |
| Model | qwen3.8-27b |
Under Model Configuration:
| Setting | qwen3.8-27b |
qwen3-coder-30b |
|---|---|---|
| Context Window | 262144 | 32768 |
| Max Output Tokens | 8192 | 8192 |
| Image Support | off | off |
| Computer Use | off | off |
Input Price and Output Price are optional. What each request cost is in the dashboard.
Which model
Roo Code works only through native tool calling: a model without tools cannot
run it. On 11 October 2026 these models returned tool calls, with and without
streaming: qwen3.8-27b, qwen3-coder-30b, qwen3.6-35b and gpt-oss-120b.
apertus-70b-instruct does not take tools.
Pick qwen3.8-27b for most work: Roo Code's instructions, its tool
descriptions and the files it reads all go into the context, and this model
has 262,144 tokens of it. qwen3-coder-30b is faster for small tasks that fit
in 32,768 tokens. More in Which model.
Limits
- 8,192 output tokens per answer, the reasoning of
qwen3.8-27bincluded. Long files are safer written in parts. - No reasoning to show: the API does not return it.