Tencent Hunyuan Hy3: open 295B MoE with 21B active under Apache 2.0
Tencent open-sourced Hy3, a 295B-parameter mixture-of-experts model with only 21B parameters active per token, under Apache 2.0 with free commercial use. 256K context, 192 experts with top-8 routing, MTP layer for faster decoding, SWE-Bench Verified 78.0, plus a separate FP8 checkpoint; weights live on Hugging Face, ModelScope, and GitHub.
Why it matters
A top-tier open model under a truly permissive license: 21B active parameters keep inference cheap while Tencent's numbers put it at 78.0 on SWE-Bench Verified and ahead of GLM-5.1 in blind expert evals on real workflows. Apache 2.0 means anyone can ship products on it.
What you could build with it
A small team selling an AI coding assistant could swap its default API model for self-hosted Hy3 weights: a 21B-active MoE that posts 78.0 on SWE-Bench Verified would cut inference spend while keeping customer code inside their own VPC, with Apache 2.0 covering commercial use.
Does it hold up?
Too early to judge — weights went public hours ago; vendor numbers (78.0 SWE-Bench Verified, 2.67/4 in blind expert evals) look strong, but independent self-host reports have not appeared yet.
Built with Tencent Hunyuan Hy3
- [SMM Flash] Tencent Hunyuan Hy3 Open Source: 295B MoE, Apache 2.0, Free for Commercial UseSMM / news.metal.com · Launch-day report: 295B MoE, 21B active, 256K context, Apache 2.0 free for commercial use; weights on HuggingFace/ModelScope/GitHub; agent and coding scores up 30% over the April preview.
- Tencent Releases Hy3: 295B Open MoE Model Under Apache 2.0NYU Shanghai RITS · Deep dive on what changed since the preview: feedback from 50+ products running Hy3 Preview, hallucination rate down from 12.5% to 5.4%, blind expert evals scoring 2.67/4 ahead of GLM-5.1 in frontend development and CI/CD.
- Tencent-Hunyuan/Hy3github · 671 ★ · Official Hy3 weights, inference recipes, and the FP8 checkpoint (plus the earlier Hy3-preview repo).
Learn more
First spotted on article: source.
More AI models
OpenAI publishes 722 AI-generated math manuscripts from an unreleased modelOpenAI released 722 mathematical manuscripts in 372 result families to GitHub under Apache-2.0, produced by an…models · JEV 0.74NVIDIA Isaac Lab 3.0 GA + Isaac OS open runtime + Agile ONE SN humanoidMajor Isaac platform expansion at GTC Berlin Oct 21: Isaac Lab 3.0 is generally available as the open-source robot…models · JEV 0.73StepFun Step-5-Preview — open BF16 weights releasedStepFun published the full BF16 weights for its flagship Step-5-Preview agentic model (600B-parameter sparse MoE, 27B…models · JEV 0.73NVIDIA Nemotron 3.5 Retriever collection: Embed 8B/1B, Rerank 8B, Parse 631MNVIDIA's Oct 21 collection drop for agentic retrieval: Nemotron-3.5-Embed in 8B BF16, 1B BF16 and 1B NVFP4…models · JEV 0.72Kimi K3.1 Pro: Moonshot's 2.8T open-weight flagship officially launchesMoonshot AI officially launched Kimi K3.1 Pro on Oct 23, a mixture-of-experts model with 2.8 trillion total / 104…models · JEV 0.72ElevenLabs Eleven v4 + v4 Turbo: new emotive TTS architecture with a 100ms real-time variantElevenLabs released Eleven v4 and Eleven v4 Turbo on September 28: a new text-to-speech architecture with inline…models · JEV 0.72
Get the week's best AI launches, plus 3 ideas worth building
One email every Saturday. Ranked by traction, not hype. Free.