oktayd/Qwen3.6-35B-v2-MoE-Ablit-Heretic-Uncensor-Hermes-MTP-Vision-FT

TEXT GENERATIONPricing:Input $0.4 / Cached $0.07 / Output $4Concurrent Unit Cost:3Model Size:35.1BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Sep 4, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

oktayd/Qwen3.6-35B-v2-MoE-Ablit-Heretic-Uncensor-Hermes-MTP-Vision-FT is a 35.1 billion parameter Mixture of Experts (MoE) vision-language model based on the Qwen3.6 architecture, with approximately 3 billion active parameters per token. This model is fine-tuned for a direct, sharp-witted, and business-minded personality, excelling in code generation, image understanding, and tool-enabled workflows. It integrates inherited refusal-reduction techniques and Hermes-style function calling for structured requests and agentic tasks.

Loading preview...

Model Overview

oktayd/Qwen3.6-35B-v2-MoE-Ablit-Heretic-Uncensor-Hermes-MTP-Vision-FT is a 35.1 billion parameter Mixture of Experts (MoE) model, where approximately 3 billion parameters are active per token, selecting 8 of 256 routed experts. This architecture allows for efficient computation while maintaining a large overall parameter count. The model features a native vision-language architecture, enabling it to understand and respond to image inputs.

Key Capabilities

  • Multimodal Understanding: Processes both text and image inputs, suitable for tasks like explaining screenshots, documents, or diagrams.
  • Code Generation & Debugging: Optimized for building and fixing code, drafting new code, and working through bugs.
  • Tool-Enabled Workflows: Designed for integration into applications that supply and execute tools, supporting structured requests and agentic tasks.
  • Customizable Personality: Fine-tuned for a direct, sharp-witted, and business-minded conversational style, with flexibility to adjust tone (professional, casual, blunt, playful).
  • Refusal Reduction: Incorporates inherited training stages (Ablit, Heretic, Uncensor) aimed at reducing refusal behavior.

Good For

  • Developers needing a powerful, locally deployable model for code generation and debugging.
  • Applications requiring image understanding (e.g., analyzing UI, OCR, diagrams).
  • Use cases benefiting from tool-calling and agentic workflows.
  • Users seeking a model with a distinct, opinionated, and adaptable personality for brainstorming and creative tasks.

This model is available in various quantized editions (BF16, GGUF, Ollama) to suit different hardware configurations, from large servers to laptops.

Finetunes
1 models