Merlin-Research/Qwen3.5-4B-Safety-Thinking

VISIONConcurrent Unit Cost:1Model Size:4.5BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Mar 2, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

Qwen3.5-4B-Safety-Thinking is a 4-billion parameter language model developed by Merlin Research, based on the Qwen/Qwen3.5-4B architecture. It is specifically optimized for structured reasoning, strict instruction adherence, and safety-aligned behavior, with a potential context length of 1M tokens. This model excels in applications requiring robust, predictable, and safe autonomous behavior, leveraging a post-training stack focused on reliable real-world performance. Its core strength lies in enhancing safety reasoning and maximizing controllability for assistant and agentic workflows.

Loading preview...

Model Overview

Merlin-Research/Qwen3.5-4B-Safety-Thinking is a 4-billion parameter model built upon the Qwen/Qwen3.5-4B base, developed by Merlin Research. It is distinguished by its rigorous post-training optimization for safety and reasoning, with a potential context length of 1 million tokens. The model leverages a LoRA-based Supervised Fine-Tuning (SFT) approach, resulting in a merged BF16 checkpoint.

Key Capabilities

  • Structured Reasoning Quality: Enhanced ability to perform step-by-step thinking, explicitly supporting the <think>...</think> format to separate reasoning from final output.
  • Instruction Adherence: Superior capability to follow complex and strict guidelines provided in prompts.
  • Safety-Aligned Behavior: Designed for safe operation in practical assistant and autonomous agent workflows, with increased robustness against misalignment and adversarial inputs.
  • Controllability: Optimized for maximizing control over model responses and ensuring consistency.

Training and Data

The model was trained on Merlin Research's private datasets, focusing on reasoning reliability, instruction-following robustness, safety behavior refinement, and misalignment reduction. It incorporates Anthropic’s Bloom&Petri framework for behavioral alignment.

Intended Use Cases

This model is particularly well-suited for:

  • Building safety-oriented reasoning assistants and chatbots.
  • Tasks demanding strict, constrained instruction-following.
  • AI alignment, safety research, and robustness testing.
  • Agentic workflows requiring predictable and safe autonomous behavior.