Qwen3.8-27B, served from Cape Town
Cape Inference runs the open Qwen3.8-27B model on our own machines in Cape Town, South Africa. We provide it to developers through OpenRouter.
Using it
Call the model qwen/qwen3.8-27b on OpenRouter.
OpenRouter chooses a provider for each request, and its provider settings let you ask for Cape Inference by
name. We don't sell access directly.
What we run
- Model
- Qwen3.8-27B by the Qwen team, Apache 2.0 licence
- Weights
- 4-bit (NVFP4)
- Context
- 131,072 tokens, with answers up to 32,768 tokens
- Features
- Streaming, tool calls, JSON output with schemas, reasoning on or off, log probabilities
- Hardware
- NVIDIA GB10 machines in Cape Town
Your data
We don't store or log prompts or answers, and we never train on them. The privacy policy has the details.