trohrbaugh/Qwen3.5-9B-heretic-v2
The trohrbaugh/Qwen3.5-9B-heretic-v2 is a 9 billion parameter causal language model, a decensored version of Qwen/Qwen3.5-9B, developed using the Heretic v1.2.0 tool. This model features a unified vision-language foundation, an efficient hybrid architecture, and supports a native context length of 262,144 tokens, extensible up to 1,010,000 tokens. It is optimized for multimodal tasks, including reasoning, coding, agentic capabilities, and visual understanding, while offering reduced refusal rates compared to its original counterpart.
Loading preview...
Model Overview
This model, trohrbaugh/Qwen3.5-9B-heretic-v2, is a 9 billion parameter decensored variant of the Qwen/Qwen3.5-9B model, created using the Heretic v1.2.0 tool. It maintains the advanced capabilities of the original Qwen3.5, including a unified vision-language foundation and an efficient hybrid architecture combining Gated Delta Networks with sparse Mixture-of-Experts for high-throughput inference. The model supports a substantial native context length of 262,144 tokens, which can be extended up to 1,010,000 tokens using YaRN scaling techniques.
Key Differentiators
- Decensored Output: Significantly reduced refusal rates (6/100) compared to the original model (100/100), offering more direct responses.
- Multimodal Capabilities: Excels in vision-language tasks, including STEM, puzzle-solving, general VQA, text recognition, document understanding, spatial intelligence, and video understanding.
- Agentic Features: Strong performance in general agent benchmarks (e.g., BFCL-V4 at 66.1, TAU2-Bench at 79.1) and visual agent tasks (e.g., OSWorld-Verified at 41.8).
- Tool Calling: Demonstrates enhanced tool calling capabilities, scoring 45.6/31.9 on TIR-Bench and 90.1/88.5 on V*.
- Global Linguistic Coverage: Expanded support for 201 languages and dialects, enabling broad international deployment.
Recommended Use Cases
- Applications requiring uncensored or less restricted responses: Ideal for scenarios where the original model's refusal rates are prohibitive.
- Complex multimodal tasks: Suitable for integrating text, image, and video inputs for advanced reasoning, coding, and agentic workflows.
- Long-context processing: Effective for tasks requiring extensive context, such as document analysis or detailed code comprehension, with its extensible context window.
- Multilingual deployments: Can be leveraged for applications needing broad language support and nuanced cultural understanding.