MODEL CATALOG

Models. Full disclosure.

Text and vision models with pinned artifacts, known hardware, and explicit capability limits.

Private beta · text

Qwen3.8-27B Abliterated

qwen3.8-27b-abliterated

AWQ W4A16 · NVIDIA RTX 5090 · 131,072 total tokens · completion within remaining context.

Non-thinking by default. Text, SSE streaming, and tools are on. Thinking mode remains unsupported. For images, use multimodal Chat Completions (data-URL image_url); the gateway routes those to the vision model.

Qwen model family; community build by twolven. Independently served by AblitAI, with no affiliation or endorsement by Alibaba or Qwen.

Read the model card ↗

Private beta · vision

Qwen3-VL-8B Abliterated

Huihui-Qwen3-VL-8B-Instruct-abliterated

Multimodal chat · NVIDIA RTX 3090 · ~8K context on the vision sidecar · inline data-URL images.

Any Chat Completions request that includes an image_url content part is routed here automatically. Use data:image/...;base64,... URLs only. Remote http(s) image URLs, audio, and video are rejected.

Community abliterated Qwen3-VL checkpoint, independently served by AblitAI.

Read the model card ↗

Choose by the current envelope

Base-model capabilities and repository names do not automatically describe these endpoints. Read the client support matrix, the API quickstart (including vision), and current pack pricing before integrating. Live status reports availability separately from the model specification.