[

](https://ko-fi.com/abcuo)

Fine-tuned from Kizzington/Qwen3-VL-8B-Thinking-heretic via Unsloth GRPO. Vision-Language model (Qwen3-VL 8B backbone). Trained on outputs from: GLM5 / Opus 4.6 / Sonnet 4.5 / Kimi / Grok / Gemini-3-pro.

Benchmarks run with lm-evaluation-harness. Text backbone extracted from VL model for evaluation compatibility.