xtools-at/Qwen3.5-9B-Claude-3xMerge

VISIONPricing:Input $0.431 / Cached $0.0862 / Output $1.12Concurrent Unit Cost:1Model Size:9BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Oct 13, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

xtools-at/Qwen3.5-9B-Claude-3xMerge is a 9 billion parameter experimental language model based on the Qwen3.5-9B architecture, created by xtools-at. This model is a weighted merge of three fine-tuned variants: Qwythos-9B-v2, Qwen3.5-9B-Fable-5-v1, and Qwopus3.5-9B-Coder. It is designed to combine general-purpose reasoning, conversational generation, and coding capabilities, including support for agent workflows and tool use, making it suitable for diverse analytical and development tasks.

Loading preview...

Qwen3.5-9B Claude Mythos/Fable/Opus Triple-Merge

This model is an experimental 9 billion parameter weighted merge of three distinct Qwen3.5-9B fine-tunes, aiming to consolidate diverse capabilities into a single compact model. It combines reasoning-oriented generation, conversational behavior, and coding-focused agent workflows. The merge uses a linear weighted average of:

  • Qwythos-9B-v2: Emphasizes reasoning, stable generation, and reduced repetitive output, incorporating Final-Token Preference Optimization (FTPO) to mitigate looping behavior.
  • Qwen3.5-9B-Fable-5-v1: Fine-tuned on Claude Fable 5 / Claude Code traces, contributing conversational and instruction-following trajectories.
  • Qwopus3.5-9B-Coder: Focuses on coding, structured reasoning, function calling, and agentic tool-use workflows, trained on datasets like hermes-agent-reasoning-traces and Claude Opus-style trace-inversion data.

Key Capabilities

  • Integrated Reasoning: Combines general analytical reasoning with structured problem-solving.
  • Conversational Generation: Enhanced dialogue and instruction-following abilities from Fable 5 fine-tuning.
  • Coding & Agent Workflows: Supports code generation, debugging, function calling, and tool-assisted tasks.
  • Repetition Mitigation: Incorporates techniques from Qwythos v2 to reduce repetitive output.

Good For

  • Experimentation: Ideal for exploring the synergy of reasoning, conversation, and coding in a single model.
  • Multi-task Applications: Suitable for use cases requiring a blend of analytical, generative, and development-oriented functions.
  • Agent Development: Provides a foundation for building and testing agentic systems with tool-use capabilities.
  • Long-Context Tasks: Designed to support tasks involving large prompts, though actual performance requires independent validation.

Finetunes
2 models