Site
About this guide
Last updated
This is an unofficial field guide to Hy4 Flash. It is not affiliated with, endorsed by, or published by Tencent. The model is theirs. The pages here are a reading of their docs.
How facts are checked
Pages were written against primary sources on 21 November 2026. A figure appears only when one of those sources states it. The sources that carry the load are the Hy4 Flash model card, the GitHub repository and its LICENSE file, Tencent's 16 November 2026 announcement, the vLLM recipe, AngelSlim's GGUF card and hardware notes, the tencent/Hy4 model card, and Tencent Cloud's TokenHub API and pricing pages.
File sizes for the three Hugging Face checkpoints are sums of the tree API on that date. GGUF sizes are the figures on AngelSlim's card, checked against the listed file bytes. The llama.cpp pull request AngelSlim cites, number 32517, was open and unmerged that day. If it merges, or if Tencent moves a Docker flag, the card wins and this site should be corrected.
What this site will not invent
- Benchmark scores. WorkIndex, the blind study, and the base-model table are quoted as Tencent's, with their caveats.
- Prices. Token amounts are the list prices on Tencent's pricing pages, including the discount window those pages print.
- A minimum VRAM number. The recipe's 529 GB and 345 GB hints are described the way the recipe describes them: download size times 1.2, not a measured minimum.
- A laptop quant. AngelSlim's smallest official pair is 125.73 GiB, and their speed tables name a 128 GB Mac and a 128 GB Spark.
- A claim that a GPU rental or an API router has a one-click Hy4 Flash button, unless their own page says so.
Affiliate links
Some links are affiliate links; we may earn a commission at no extra cost to you. Those links are only on the cloud page. Other pages link to that page when a rented GPU is the practical next step. The disclosure is repeated in the privacy policy.
Corrections
Send a correction to hello@hy4flash.com. Model bugs and license requests belong to Tencent. The model card lists hunyuan_opensource@tencent.com for research and product questions, and the license lists hy_partners@tencent.com for Section 10.
Sources
- tencent/Hy4-flash on Hugging Face Model card, architecture table, deployment commands, and license pointer.
- Tencent, “Tencent Launches Hy4 flash” Announcement dated 16 November 2026.
- Tencent-Hunyuan/Hy4-flash on GitHub Repository created 17 November 2026.
- vLLM recipe for tencent/Hy4-flash
- AngelSlim/Hy4-flash-GGUF Official GGUF builds linked from the model card.
- Tencent Cloud, TokenHub model pricing English list prices. Page updated 19 November 2026.
- Tencent Hy License, Version 1.0 Apache License 2.0 terms plus Section 10 for large MaaS operators.