KU-DFI/TelecomGPT-R1

VISIONConcurrency Cost:2Model Size:27BQuant:FP8Ctx Length:32kTool Calling:SupportedPublished:May 18, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Cold

KU-DFI/TelecomGPT-R1 is a 27 billion parameter open model developed by KU/DFI, specifically designed for universal reasoning across diverse telecommunications tasks. It achieves state-of-the-art performance on the GSMA Open Telco Leaderboard with an 89.6% average score, outperforming both general-purpose and telecom-specialized models, including larger closed-source frontier LLMs. This model excels in protocol understanding, knowledge QA, modeling & computation, and fault analysis, making it ideal for self-hosted, auditable telecom AI applications.

Loading preview...

TelecomGPT-R1: Specialized Telecom Reasoning

TelecomGPT-R1 is a 27 billion parameter open-source model developed by KU/DFI, engineered for universal reasoning within the telecommunications domain. It has achieved a state-of-the-art average score of 89.6% on the GSMA Open Telco Leaderboard, surpassing all other models, including significantly larger open-source and closed-source general-purpose LLMs, as well as other telecom-specialized models.

Key Capabilities

  • Comprehensive Telecom Reasoning: Excels across four critical axes: protocol understanding (3GPP/O-RAN), knowledge QA (vendor/operator facts), modeling & computation (RF/queueing derivations), and fault analysis (RAN drive-test logs).
  • Parameter Efficiency: Achieves top performance with approximately 25x fewer active parameters than the next-best open entrant, demonstrating significant efficiency.
  • Open-Source Advantage: Designed for self-hosting behind operator firewalls, enabling direct use on confidential data, fine-tuning on proprietary subsystems, and auditing for compliance.
  • Unified Policy: Trained on a single, unified telecom-reasoning corpus (158,915 examples) covering heterogeneous modalities, allowing for seamless cross-modal reasoning.

Good For

  • Telecom Operators & Vendors: Ideal for applications requiring deep, verifiable reasoning on telecom-specific data, such as diagnosing network issues, interpreting standards, and performing complex computations.
  • Research & Development: Provides an open foundation model for further innovation in telecom AI, allowing for auditing, improvement, and adaptation by the wider industry.
  • Edge/Device-Side Inference: Future smaller variants are planned to extend deployment from data-center GPUs down to on-device telecom intelligence.

Finetunes
1 models