Model WarsJuly 6, 2026via The Decoder
Tencent releases Hy3 open-source model that allegedly matches models up to five times its active size
Why it matters
Tencent's MoE architecture challenges the scaling assumption that bigger always means better. If validated, this efficiency claim reshapes how enterprises approach model deployment costs and performance trade-offs.
Key signals
- Tencent releases Hy3 open-source LLM
- 295B total parameters, 21B active (MoE architecture)
- Claims performance parity with models 2-5x larger
- Hallucination rate: 5.4% (stated 50% reduction)
- Open-source release
- Published July 6, 2026
The hook
295B parameters. 21B active. Tencent's Hy3 claims to match models 5x larger while cutting hallucinations in half.
Tencent has released Hy3, an open-source language model with 295 billion parameters built on a mixture-of-experts architecture. Only 21 billion parameters are active at any given time. Tencent says Hy3 matches models two to five times its size while cutting its hallucination rate in half to 5.4 percent.