leandronunes/qwen3.8-4b-a1-dare-ties

VISIONPricing:Input $0.4 / Cached $0.08 / Output $0.8Concurrent Unit Cost:1Model Size:4.5BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Sep 22, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

The leandronunes/qwen3.8-4b-a1-dare-ties model is a 4.5 billion parameter language model, merged from Qwen3.5-4B, Qwen3.8-4B-Distill, and Agents-A1-4B using the DARE-TIES method. It is designed to combine strong logical deduction capabilities with agentic reasoning, making it suitable for complex problem-solving tasks. This model also features a multimodal vision tower and a hybrid attention architecture, supporting a substantial context length of 262,144 tokens.

Loading preview...

Model Overview

leandronunes/qwen3.8-4b-a1-dare-ties is a 4.5 billion parameter language model created by leandronunes through a DARE-TIES merge. It combines the strengths of Qwen3.8-4B-Distill (known for logical deduction) and InternScience/Agents-A1-4B (for agentic capabilities), built upon the Qwen3.5-4B base.

Key Capabilities & Features

  • Enhanced Reasoning: Aims to integrate strong logical deduction (from Qwen3.8-4B-Distill's 78% on BBH Logical Deduction) with agentic reasoning (from Agents-A1-4B's 38% on BBH Causal Judgement).
  • Multimodal (VLM): Includes a 24-block vision tower in addition to the language model, enabling visual understanding.
  • Hybrid Attention Architecture: Utilizes a unique structure of 3 linear_attention (Gated DeltaNet) layers followed by 1 full_attention, repeated 8 times.
  • Extended Context Window: Supports a very large context length of 262,144 tokens (256K).
  • Configurable Thinking Mode: Offers an enable_thinking option to control extended reasoning behavior.

Why Choose This Model?

This model is particularly well-suited for applications requiring a blend of:

  • Complex Problem Solving: Where both step-by-step logical deduction and high-level agentic planning are crucial.
  • Multimodal Understanding: For tasks that involve processing both text and visual inputs.
  • Long Context Processing: Its 256K context window makes it ideal for handling extensive documents or conversations.

While specific benchmark results for the merged model are pending, its design targets improved performance in areas demanding sophisticated reasoning.

Finetunes
1 models