MODEL
Model
The model identifier in your IDE config is a brand name, not the name of the underlying weight. This page states, in plain terms backed by the running configuration, what devpipe-code-1 actually is right now.
01What runs behind the name today
| Property | Value |
|---|---|
| Public model name (use this in your IDE config) | devpipe-code-1 |
| Also answered under (legacy names, kept forever) | qwen-3.8-27b, qwen-27b, gpt-4o, claude-3-5-sonnet-20241022 |
| Backend weight actually serving requests | Qwen/Qwen3.8-27B-Instruct-AWQ |
| Context window | 32,768 tokens |
| Weight quantization | AWQ |
| In service since | Not yet in service |
| Public names retired before this one | None — this is the first model shipped under a public name. |
02How a model change is handled
devpipe-code-1 is a brand name kept stable across model changes on purpose: IDE and CLI configuration files belong to the customer, and dev.pipe cannot reach into them. When the backend weight changes, the name in your config keeps working because it is remapped in one place; every name this service has ever shipped stays routable for exactly the same reason.
What this service cannot offer is a frozen snapshot name that keeps serving the retired model indefinitely. The cluster is four GPUs sized for one model generation; running two generations side by side would halve throughput for everyone. A model change replaces the weight behind the name for all customers at once.
⚠ MISSING — OPERATOR INPUT REQUIRED
Advance notice before a model change
How many days a model change is announced in advance is a business decision the operator has not made yet. Until it is, no notice period is promised here.