tepirale/Ornith-Agents-A1-3.6-35B-A3B-dare_ties

Hugging Face
TEXT GENERATIONConcurrent Unit Cost:3Model Size:35.1BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Jul 3, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Warm

tepirale/Ornith-Agents-A1-3.6-35B-A3B-dare_ties is a 35 billion parameter language model created by tepirale, merged using the DARE TIES method with Qwen/Qwen3.5-35B-A3B as its base. It integrates capabilities from deepreinforce-ai/Ornith-1.0-35B and InternScience/Agents-A1, excelling in reasoning, tool-calling, and general language understanding. This model is optimized for complex tasks requiring structured thought processes and interaction with external tools, demonstrating strong performance across various benchmarks including MMLU, HumanEval, and MathQA.

Loading preview...

Model Overview

tepirale/Ornith-Agents-A1-3.6-35B-A3B-dare_ties is a 35 billion parameter language model resulting from a DARE TIES merge. It combines the strengths of Qwen/Qwen3.5-35B-A3B as its base with deepreinforce-ai/Ornith-1.0-35B and InternScience/Agents-A1. This merge aims to enhance the model's capabilities in complex reasoning and agentic behaviors.

Key Capabilities

  • Advanced Reasoning: The model is designed with a strong emphasis on reasoning, utilizing a <think>...</think> mechanism for structured thought processes, as indicated by its reasoning-parser qwen3 configuration.
  • Tool-Calling: It supports sophisticated tool-calling functionalities, configured with tool-call-parser qwen3_xml, making it suitable for agent-based applications that require interaction with external tools.
  • Multimodal Support: The model can handle multimodal inputs, specifically images, with a limit of 4 images per prompt.
  • Strong Benchmark Performance: Achieves competitive scores across a range of benchmarks, including:
    • MMLU: 82.2
    • HellaSwag: 92.2
    • HumanEval: 92.7
    • MathQA: 77.6

Ideal Use Cases

  • Agentic AI Systems: Its robust tool-calling and reasoning capabilities make it well-suited for developing AI agents that can plan, execute, and interact with environments.
  • Complex Problem Solving: Excels in tasks requiring multi-step reasoning, logical deduction, and structured output.
  • Code Generation and Understanding: Demonstrated strong performance on coding benchmarks like HumanEval and MBPP, suggesting proficiency in programming-related tasks.
  • Multimodal Applications: Can be utilized in scenarios where understanding and responding to image inputs is crucial.

Finetunes
1 models