AIOR-Research/SOLID-StepORLM

TEXT GENERATIONPricing:Input $0.468 / Output $1.82Concurrent Unit Cost:1Model Size:8BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Sep 8, 2026Architecture:Transformer Featherless Exclusive Cold

AIOR-Research/SOLID-StepORLM is an 8 billion parameter language model built from Chenyu-Zhou/StepORLM-Qwen3-8B, specifically designed for operations research modeling and solver-backed answer generation. It was trained using GRPO and solver-informed token-level KL supervision, utilizing a COPT-style StepORLM response template. This model excels at generating optimization code and solutions for complex operations research problems, as demonstrated by its performance on OptMATH, MAMO-Complex, and InOR datasets. It is optimized for tasks requiring the formulation and solution of mathematical optimization problems.

Loading preview...

AIOR-Research/SOLID-StepORLM: Operations Research Language Model

AIOR-Research/SOLID-StepORLM is an 8 billion parameter model derived from Chenyu-Zhou/StepORLM-Qwen3-8B, specifically engineered for operations research (OR) modeling and generating solver-backed solutions. The model incorporates SOLID (Solver-Informed Self-Distillation), a training methodology that leverages GRPO and token-level KL supervision informed by solver outputs.

Key Capabilities

  • Operations Research Modeling: Generates optimization models and solutions using a COPT-style StepORLM response template.
  • Solver-Backed Answer Generation: Designed to produce answers that can be verified and executed by compatible COPT environments.
  • Specialized Training: Utilizes GRPO and solver-informed self-distillation for enhanced performance in OR tasks.

Performance Highlights

Evaluated on standard operations research datasets, the model demonstrates strong capabilities:

  • OptMATH: Achieves maj@64 of 31.33 and pass@1 of 18.25.
  • MAMO-Complex: Shows robust performance with maj@64 of 70.44 and pass@1 of 66.43.
  • InOR: Records maj@64 of 48.00 and pass@1 of 39.81.

Good For

  • Developers and researchers working on operations research problems requiring automated model generation.
  • Applications that need solver-compatible optimization code output.
  • Tasks involving mathematical optimization and decision-making support.

Finetunes
1 models