Hy3 is a large language model released by Tencent, released under the apache-2.0 license, with a claimed knowledge cutoff of 2026-07-06. Its weights, configuration, and tokenizer files could not be hashed successfully -- see the error below by Weight Registry.
Description
Hy3 is a 295B-parameter Mixture-of-Experts model from Tencent's Hunyuan team, released April 23, 2026, with 21B active parameters, 192 experts with top-8 routing, and up to 256K context. It's a hybrid fast/slow-thinking model built for reasoning, agentic workflows, and production use, with a configurable reasoning-effort setting.
Metadata
Hugging Face stats
101,737 downloads on Hugging Face · 971 likes · pipeline: text-generation · library: transformers
Weight Hashes
Hashing failed -- see the note below for the error.
Error
('Connection broken: IncompleteRead(7194057243 bytes read, 53771765 more expected)', IncompleteRead(7194057243 bytes read, 53771765 more expected))