Tencent open-sources Hy3 295B MoE AI model under Apache 2.0

Tencent's Hunyuan team has released the full version of Hy3, a 295-billion-parameter Mixture-of-Experts artificial intelligence model with 21 billion active parameters, under the Apache 2.0 license, VentureBeat reported Monday. The release reverses April's preview terms. VentureBeat said Apache 2.0 carries no regional exclusions and no field-of-use restrictions. Tencent said the model will be free on OpenRouter for two weeks. The architecture is unchanged from the preview, according to the report: 295 billion total parameters, 21 billion active per forward pass via top-8 routing across 192 experts, a 3.8 billion-parameter multi-token prediction layer for speculative decoding, and a 256K context window. After a late-April preview, the team collected feedback from more than 50 product teams, fixed issues in task execution and interaction, and scaled up post-training, according to the model card cited by VentureBeat. Tencent positions the full release as outperforming similar-size models and rivaling flagship open-source models with two to five times the parameters. In a company-run blind test with 270 experts and 312 valid comparisons, Tencent reported Hy3 scored 2.67 out of 4 against GLM-5.1's 2.51. VentureBeat said nearly all competitor figures in Tencent's appendix came from Tencent's own runs and that independent verification was still pending as of publication. Tencent said internal evaluations showed the hallucination rate fell from 12.5% to 5.4% versus the preview. VentureBeat said the weights are available on Hugging Face.