Merlin-Research/Pluto
Merlin-Research/Pluto is a 9 billion parameter coding and reasoning model developed by Merlin Research, built on Qwen/Qwen3.5-9B-Base. It features a precision-first design, a 1,000,000 token context window, and is optimized for agentic coding environments. Notably, it incorporates Adaptive Entropy Regularization with quantum noise from IBM Quantum Kingston hardware during training, making it robust for complex code generation and technical reasoning tasks.
Loading preview...
Pluto: Precision Coding and Reasoning Model
Pluto is a 9 billion parameter model from Merlin Research, specifically engineered for precision and robustness in coding and reasoning tasks. Built upon Qwen/Qwen3.5-9B-Base, its training prioritizes minimizing errors over maximizing fluency, making it highly effective where code correctness is critical.
Key Capabilities
- 1 Million Token Context: Handles entire repositories or multi-file diffs, maintaining coherent reasoning across vast codebases without chunking.
- Agentic Deployment Ready: Fine-tuned for seamless integration with Claude Code, OpenAI Codex/Assistants API, and local large-codebase workflows, supporting compatible message structures and function calling.
- Quantum Entropy Regularization (AER): Utilizes quantum noise from IBM Quantum Kingston hardware during RL training to enhance robustness on out-of-distribution inputs and prevent entropy collapse.
- Distillation from Frontier Models: Incorporates knowledge distillation from advanced coding models and a private dataset of reasoning traces to achieve deep reasoning capabilities at a 9B scale.
Good For
- Complex code generation and refactoring
- Multi-file codebase analysis and bug detection
- Agentic coding pipelines and code review
- Architecture planning and technical reasoning
- Local deployment with large private codebases