Tencent Cloud
Hy3-preview is described by Tencent Hunyuan as a 295B-parameter Mixture-of-Experts model with 21B active parameters, 3.8B MTP layer parameters and 256K context length. The English README says it improves on complex reasoning, instruction following, context learning, coding and agent tasks, and provides vLLM and SGLang deployment examples with an OpenAI-compatible local API.
Editorial verdict
Teams evaluating open Chinese reasoning and agent models with long-context deployment requirements.
Avoid it if you need a small model or a fully managed hosted API without infrastructure work.
Hy3-preview is the English-documented Hunyuan open LLM profile.
Open weights are free to download; self-hosting requires large GPU infrastructure
Hugging Face, ModelScope, GitCode, Self-hosted GPU infrastructure
Commercial use should follow the current product, API, model license and billing terms.
Review prompt, file, media upload, retention and training-use terms before sensitive workloads.
Use it when open weights, long context and local serving are more important than hosted convenience.
Tencent specifically calls out SWE-bench, Terminal-Bench, BrowseComp, WideSearch, ClawEval and WildClawBench-style scenarios.
Model names, quotas, release status, regional access and commercial terms can change quickly; recheck official sources before procurement or production use.
mimo-v2-flash
deepseek-v4-api
qwen
other · en · verified 2026-05-17
Confirms model size, context, deployment, model links and license.
official · en · verified 2026-08-04
Official Hunyuan research site referenced by the open-source organization.
Last checked: 2026-08-04