arXiv:2512.10430v1 Announce Type: new 
Abstract: We introduce T-pro 2.0, an open-weight Russian LLM for hybrid reasoning and efficient inference. The model supports direct answering and reasoning-trace generation, using a Cyrillic-dense tokenizer and an adapted EAGLE speculative-decoding pipeline to reduce latency. To enable reproducible and extensible research, we release the model weights, the T-Wix 500k instruction corpus, the T-Math reasoning benchmark, and the EAGLE weights on Hugging Face. These resources allow users to study Russian-language reasoning and to extend or adapt both the model and the inference pipeline. A public web demo exposes reasoning and non-reasoning modes and illustrates the speedups achieved by our inference stack across domains. T-pro 2.0 thus serves as an accessible open system for building and evaluating efficient, practical Russian LLM applications.

تم تقديم T-pro 2.0 كنموذج لغة روسية مفتوح الوزن مصمم للتفكير الهجين والاستدلال الفعال، مع وجود مُرمز كثيف بالحروف السيريلية وخط أنابيب فك تشفير استباقي مُعدل EAGLE لتحسين الأداء. تتوفر موارد النموذج، بما في ذلك الأوزان والمعايير، على Hugging Face، مما يسهل البحث في التفكير باللغة الروسية.

Se ha presentado T-pro 2.0 como un modelo de lenguaje ruso de pesos abiertos diseñado para el razonamiento híbrido y la inferencia eficiente, con un tokenizador denso en cirílico y un pipeline de decodificación especulativa EAGLE adaptado para mejorar el rendimiento. Los recursos del modelo, incluidos los pesos y los benchmarks, están disponibles en Hugging Face, facilitando la investigación en razonamiento en lengua rusa.

T-pro 2.0 a été introduit comme un modèle de langue russe à poids ouvert conçu pour le raisonnement hybride et l'inférence efficace, avec un tokenizer dense en cyrillique et un pipeline de décodage spéculatif EAGLE adapté pour améliorer les performances. Les ressources du modèle, y compris les poids et les benchmarks, sont disponibles sur Hugging Face, facilitant la recherche en raisonnement en langue russe.

T-pro 2.0 has been introduced as an open-weight Russian language model designed for hybrid reasoning and efficient inference, featuring a Cyrillic-dense tokenizer and an adapted EAGLE speculative-decoding pipeline to enhance performance. The model's resources, including weights and benchmarks, are available on Hugging Face, facilitating research in Russian-language reasoning.

T-pro 2.0: An Efficient Russian Hybrid-Reasoning Model and Playground

One More Thing in AI – Your Shortcut to AI Mastery

T-pro 2.0: An Efficient Russian Hybrid-Reasoning Model and Playground

Was this article worth reading? Share it

One More Thing in AI

LucidQuery AI

Airparser

Fakeface

GPTHumanizer

Zemith-3bda3b

Ready to build your own newsroom?