roskosmos19/Rhea-4B-Coding

TEXT GENERATIONConcurrency Cost:1Model Size:4BQuant:BF16Ctx Length:32kTool Calling:SupportedPublished:Jun 17, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Cold

Rhea-4B-Coding by Roskosmos19 is a 4 billion parameter language model, an optimized version of Aquiles-ai/Athenea-4B-Coding, specialized in code reasoning, debugging, and multi-pass problem solving. This model is designed for detailed 3-pass reasoning in software development, algorithm design, and code comprehension tasks, utilizing explicit reasoning traces and agentic tools. It excels at iterative code refinement and logical problem-solving, offering an uncensored output generation for research and experimentation.

Loading preview...

Rhea-4B-Coding: Multi-Pass Code Reasoning Model

Rhea-4B-Coding is a 4 billion parameter model developed by Roskosmos19, building upon the Aquiles-ai/Athenea-4B-Coding base. It is specifically optimized for advanced code reasoning, debugging, and multi-pass problem-solving in software development. The model employs a unique 3-pass reasoning architecture (First implementation, Self-review, Final optimized version) guided by special tokens like <think>, <review>, and <final> to ensure iterative refinement and improved logical consistency.

Key Capabilities:

  • Multi-Pass Processing: Structured 3-step reasoning for comprehensive code development.
  • Agentic Tools: Designed to integrate with AI agents for enhanced functionality.
  • Step-by-step Reasoning: Utilizes thinking blocks for detailed thought processes.
  • Self-Review: Capable of detecting bugs, addressing edge cases, and optimizing code.
  • Uncensored Output: Provides full expressive freedom for research and experimentation.
  • Specialization: Strong performance in algorithmic tasks and debugging scenarios.

Good For:

  • Developers requiring iterative code refinement and optimization.
  • Research into agentic code generation and multi-pass reasoning systems.
  • Complex logical problem-solving and algorithm design.
  • Debugging and code comprehension tasks where detailed reasoning is beneficial.

The model was fine-tuned on the Aquiles-ai/Athenea-Coding-100k dataset, which includes diverse programming challenges and structured reasoning chains across multiple languages.

Finetunes
1 models