NitrAI/OpenGCM-v2

VISIONConcurrent Unit Cost:1Model Size:9BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Jun 27, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

NitrAI/OpenGCM-v2 is a 9 billion parameter reasoning-focused model developed by NitrAI, built upon the Qwen3.5-9B base architecture with a 262k context window. This model excels at complex coding-agent trajectories, multi-step math logic, and system-level reasoning, distilled from frontier LLMs like GPT-5.5, Claude-Fable-5, and GLM-5.2. It is optimized for efficient performance on consumer hardware, making it suitable for applications requiring strong analytical and problem-solving capabilities.

Loading preview...

Overview

NitrAI's OpenGCM-v2 is a 9 billion parameter model specifically designed for advanced reasoning tasks. It leverages the Qwen3.5-9B base model, known for its state-of-the-art architecture and a substantial 262k context window. The primary goal of OpenGCM-v2 is to distill sophisticated reasoning abilities—including complex coding-agent trajectories, multi-step mathematical logic, and system-level reasoning—from powerful frontier LLMs such as GPT-5.5, Claude-Fable-5, and GLM-5.2. This distillation process aims to create a highly efficient model that runs effectively on consumer-grade hardware.

Key Capabilities

  • Exceptional Reasoning: Demonstrates strong performance in mathematical reasoning and step-by-step logical decomposition, as evidenced by solving AIME sequence problems perfectly.
  • Code Analysis: Highly capable of localized code reasoning and bug patch verification, performing well on benchmarks like SWE-bench Pro.
  • Distilled Intelligence: Benefits from a meticulously curated dataset of 597 high-signal QA items, totaling over 900,000 tokens, distilled from advanced models to enhance its problem-solving prowess.
  • Consumer Hardware Friendly: Trained using techniques like DoRA and Unsloth's optimized kernels to ensure efficient operation on single consumer GPUs.

Good For

  • Applications requiring robust mathematical problem-solving and logical deduction.
  • Tasks involving code analysis, bug fixing, and agentic coding workflows.
  • Use cases where complex, multi-step reasoning is critical, but computational resources are limited to consumer hardware.

Limitations

  • May exhibit occasional instability or context drift during extremely long inference generations. It is recommended to use a lower temperature (e.g., 0.2 or 0.4) and structured system prompts for optimal results.