d0gra/Qwen3-4B-Thinking-2507-say-no-more

TEXT GENERATIONPricing:Input $0.4 / Cached $0.08 / Output $0.8Concurrent Unit Cost:1Model Size:4BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Oct 7, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

The d0gra/Qwen3-4B-Thinking-2507-say-no-more is a 4.0 billion parameter causal language model, a decensored version of Qwen/Qwen3-4B-Thinking-2507, optimized for complex reasoning tasks. It features a native context length of 262,144 tokens and significantly improved performance across logical reasoning, mathematics, science, coding, and academic benchmarks. This model is specifically designed for "thinking mode" applications, excelling in scenarios requiring deep analytical processing and enhanced long-context understanding.

Loading preview...

Overview

This model, d0gra/Qwen3-4B-Thinking-2507-say-no-more, is a decensored version of the original Qwen/Qwen3-4B-Thinking-2507, created using the Heretic v1.4.0 tool. It is a 4.0 billion parameter causal language model with a native context length of 262,144 tokens, specifically designed to enhance "thinking capability" for complex reasoning tasks.

Key Capabilities

  • Enhanced Reasoning: Significantly improved performance on logical reasoning, mathematics, science, coding, and academic benchmarks requiring human expertise.
  • General Capabilities: Markedly better instruction following, tool usage, text generation, and alignment with human preferences.
  • Long-Context Understanding: Features enhanced 256K long-context understanding, crucial for complex problem-solving.
  • Thinking Mode: This version exclusively supports a "thinking mode," automatically including <think> in the chat template to enforce deeper processing.
  • Reduced Refusals: Compared to the original model, this decensored version shows a substantial reduction in refusals (4/100 vs. 99/100).

Good For

  • Highly Complex Reasoning Tasks: Recommended for applications demanding deep analytical processing, such as advanced mathematical problems, scientific simulations, and intricate coding challenges.
  • Agentic Use Cases: Excels in tool calling capabilities, especially when integrated with frameworks like Qwen-Agent, which simplifies tool-calling templates and parsers.
  • Long-Context Applications: Ideal for scenarios requiring the processing and understanding of very long documents or conversations, leveraging its 262,144-token context window.
  • Benchmarking and Research: Useful for researchers and developers evaluating model performance on challenging benchmarks, particularly those focused on reasoning and problem-solving.

Finetunes
1 models