OTel: Open Telecom AI Resource with 16M Downloads and 30 Post-Trained Models
Open Telco (OTel) has been launched by researchers as an open AI resource tailored for telecommunications, aiming to overcome the shortcomings of advanced AI models in telecom-related tasks. This initiative offers derived datasets for various functions, including retrieval, reranking, instruction tuning, and safety/abstention, alongside 30 comprehensive post-trained baselines across embedding, reranking, and language models. Since its release on May 3, 2026, the models have amassed over 16 million downloads and garnered more than 157 media mentions globally. OTel enhances previous open telecom datasets and benchmarks, providing documented data sources, evaluation partitions, trained models, and safety data, ultimately improving performance in embedding retrieval to 93.5% NDCG@10 and reranking to 0.952 MRR@1, laying the groundwork for specialized telecom LLMs.
Key facts
- OTel is an open telecom AI resource with derived datasets for retrieval, reranking, instruction tuning, and safety/abstention.
- Includes 30 full-parameter post-trained baselines across embedding, reranking, and language models.
- As of May 3, 2026, models downloaded over 16 million times.
- Received 157+ pieces of media coverage worldwide.
- Provides documented telecom data sources, held-out evaluation partitions, trained embedding models, rerankers, context-grounded LLMs, and safety/abstention data.
- Post-training improves performance: embedding retrieval reaches 93.5% NDCG@10; reranking reaches 0.952 MRR@1.
- Builds on prior open telecom datasets and benchmarks.
- Aims to support intelligent networks with domain-specialized telecom LLMs.
Entities
Institutions
- arXiv