VISION MODEL DISCLOSURE
Qwen3-VL-8B Abliterated.
Multimodal image chat sidecar. The gateway routes Chat Completions that include image_url here.
Private beta. Bitcoin checkout open (beta); card closed. No uptime SLA. Vision context is smaller than the text 131K envelope.
Current serving envelope
| Stable model ID | Huihui-Qwen3-VL-8B-Instruct-abliterated |
|---|---|
| Role | Vision / multimodal sidecar (auto-routed) |
| Launch GPU | NVIDIA RTX 3090, 24GB-class VRAM |
| Serving engine | vLLM (vision) |
| Context | ~8,192 tokens on the vision sidecar |
| Image input | data:image/...;base64,... URLs only |
| Not supported | Remote http(s) image fetches; audio; video |
How routing works
Use the same /v1/chat/completions endpoint as text. If any message content part has type: "image_url", the gateway sends the request to this model. You may still pass qwen3.8-27b-abliterated as the client model id; the gateway rewrites the upstream id for the VL backend.
See the vision section in the docs for a request sketch. The text model card remains the disclosure for text-only traffic on the RTX 5090.
Origin and independence
This is a community-modified abliterated Qwen3-VL checkpoint, independently served by AblitAI. No affiliation or endorsement by Alibaba or Qwen is implied. The Acceptable Use Policy still applies.
Next: try a multimodal request or check credit packs.