logo

1. Introduction

GRM-2.6-Air-Opus is a fine-tune of LiquidAI/LFM2.5-1.2B-Thinking, trained on Claude Opus-style reasoning traces to sharpen its reasoning and boost performance on harder problems.

GRM-2.6-Air-Opus is a compact, edge-friendly model built for users who want a taste of Opus-style structured reasoning without the footprint of larger GRM models. It's the "Air" entry in the GRM-2.6 family: smaller and faster, aimed at lightweight local and on-device use.

2. Key Capabilities

  • Opus-Style Reasoning, Compact Size: Distilled from Claude Opus-style traces to produce clearer, more organized reasoning at just 1.2B parameters.
  • Edge-Ready: Built on LFM2.5-1.2B-Thinking, making it well suited for low-resource and on-device deployment.
  • Improved Problem-Solving: Targeted training on harder reasoning traces lifts performance on difficult tasks relative to the base model.

3. Performance

GRM-2.6-Air-Opus focuses on practical, lightweight reasoning: getting as much structured thinking as possible out of a 1.2B model.

Detailed Benchmarks

Benchmark GRM-2.6-Air-Opus
Knowledge & STEM
MMLU 31.99

GRM-2.6-Air-Opus is developed by OrionLLM.

Downloads last month
263
Safetensors
Model size
1B params
Tensor type
BF16
·
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for OrionLLM/GRM-2.6-Air-Opus

Finetuned
(35)
this model
Quantizations
2 models

Collection including OrionLLM/GRM-2.6-Air-Opus