OrionLLM/GRM-2.6-Plus

Hugging Face
VISIONConcurrent Unit Cost:2Model Size:27BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Apr 23, 2026License:apache-2.0Architecture:Transformer0.1K Open Weights Featherless Exclusive Warm

GRM-2.6-Plus by OrionLLM is a 27B-parameter reasoning model built on the Qwen3.6 architecture, optimized for difficult, high-complexity tasks. It focuses on structured reasoning to produce accurate, coherent, and reliable responses. This model delivers elite-level reasoning and strong performance for its size, making it suitable for advanced problem-solving, coding, and agentic applications.

Loading preview...

OrionLLM/GRM-2.6-Plus: A 27B Reasoning Model

GRM-2.6-Plus is a 27-billion parameter model developed by OrionLLM, built upon the Qwen3.6 architecture. It is specifically designed for general-purpose AI, with a strong emphasis on tackling difficult and high-complexity tasks through structured reasoning. The model aims to provide elite-level reasoning capabilities while remaining practical and efficient for advanced local and research-oriented use.

Key Capabilities

  • Elite-Level Reasoning: Optimized for complex reasoning workloads, delivering clarity, consistency, and robust step-by-step problem-solving.
  • High Performance for Size: Achieves excellent capability relative to its 27B parameter count, balancing intelligence with practical deployment.
  • Advanced Coding & Agentic Use: Well-suited for code generation, structured problem-solving, tool-style workflows, and local agentic applications.
  • Optimized for Practical Deployment: Designed to be efficient and usable on capable consumer and workstation hardware.

Performance Highlights

GRM-2.6-Plus demonstrates strong performance across various benchmarks, often outperforming models like Qwen3.6-27B and google/gemma-4-31B-it in categories such as Knowledge & STEM, Reasoning & Coding, and General Agent tasks. For instance, it scores 86.8 on MMLU-Pro, 94.2 on MMLU-Redux, 84.8 on LiveCodeBench v6, and 77.7 on SWE-bench Verified, indicating its proficiency in complex problem-solving and coding.

Good For

  • Users requiring powerful reasoning for advanced problem-solving.
  • Developers working on code generation and structured problem-solving.
  • Applications involving agentic workflows and tool-style interactions.
  • Local deployment on consumer and workstation hardware for demanding AI tasks.

Finetunes
3 models