abdur-rahman97/Dahono-4B

VISIONPricing:Input $0.4 / Cached $0.08 / Output $0.8Concurrent Unit Cost:1Model Size:4.5BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Oct 2, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

Dahono-4B is a 4.5 billion parameter sovereign legal AI model developed by Dahono Labs, built on the Qwen3.5-4B-Base architecture with a 32768-token context length. It is specifically optimized for Indonesian legal reasoning, regulatory compliance, and juridical instrument analysis, featuring integrated vision capabilities for document inspection and enhanced through Reasoning Reinforcement Learning (GRPO) with four calibrated juridical reward functions. This model excels at providing accurate legal analysis and maintaining language purity in Indonesian legal contexts.

Loading preview...

Dahono-4B: Sovereign Indonesian Legal AI with Vision

Dahono-4B is a 4.5 billion parameter large language model developed by Dahono Labs, specifically engineered for the Indonesian legal domain. Built upon the Qwen3.5-4B-Base architecture, it integrates multimodal document vision capabilities and is optimized using Reasoning Reinforcement Learning via Group Relative Policy Optimization (GRPO) with four distinct juridical reward functions. This model is designed to provide highly accurate and contextually relevant legal analysis in Bahasa Indonesia.

Key Capabilities

  • Specialized Legal Reasoning: Optimized for Indonesian law, regulatory compliance, and juridical instrument analysis, achieving an 89.40% Composite Legal Score.
  • Multimodal Document Vision: Capable of processing visual inputs (scans/photos) to validate legal document elements like agreement conditions (Pasal 1320 KUHPerdata), signature authenticity, and stamp duty validity (UU No. 10 Tahun 2020).
  • Enhanced Reasoning Purity: Features a 100% pure Indonesian language block for internal legal deductions (<think>), eliminating code-switching.
  • High Citation Precision: Achieves 92.40% precision in citing legal articles and regulations.
  • Robust Training: Trained on an expanded corpus of 6,271 samples, including 4,500 Chain-of-Thought reasoning samples and 1,771 legal glossary terms.

Good For

  • Legal Research & Document Review: Accelerating research and review of Indonesian positive law documents.
  • Regulatory Compliance Pre-checks: Performing initial checks for regulatory compliance.
  • Drafting Legal Instruments: Assisting in the formulation of initial legal drafts (e.g., summons, cooperation agreements).
  • Legal Education: Serving as an educational tool for academics and legal practitioners.
  • Integration with RAG Systems: Designed to work synergistically with Retrieval-Augmented Generation (RAG) systems for national regulations.