Tencent Opens 770B Hy4 Weights, With a $50M Revenue Cap

Tencent has published the weights of Hy4, its flagship Hunyuan model, on Hugging Face, ModelScope and GitCode. The model card puts the language backbone at 770 billion total parameters, with 49 billion active per token.

The release comes with a catch for big cloud sellers. Under the Tencent Hy License, a company that resells model access must get a separate licence if its revenue tops $50 million a year.

Important: not investment advice

This article is for information and education only. It is not investment, financial, legal or tax advice, and it is not a recommendation or an offer to buy or sell any security, ETF, option or other financial product. FourWeekMBA is not a broker-dealer or a registered investment adviser. IPO shares, leveraged and inverse ETFs, and options carry significant risk, including the loss of your entire investment; leveraged and inverse ETFs reset daily and can lose value quickly. Prices, availability, fees and eligibility change and can differ by country and broker, so check the prospectus and your broker’s terms. Figures marked “our arithmetic” are our own calculations from the cited filings and may contain errors. Do your own research and consider speaking to a licensed financial professional before making any investment decision.

Business Pill · The Model and the Harness

An evergreen Business Engineer explainer on models and the harnesses around them. It teaches the general idea only and does not discuss Hy4 or Tencent.

The key insight: Tencent published the weights of its 770-billion-parameter Hy4, with 49 billion active per token; its licence requires a separate deal for resellers above $50 million in revenue or 100 million monthly users.

What Tencent Released

Tencent’s card lists three downloads: the instruct model Hy4, an FP8 version and the pre-trained Hy4 base model. The Hugging Face listing was last updated on 20 November.

Hy4 reads text, images and video. Its context window is 1,048,576 tokens, and the card says it uses “Gated DeepSeek Sparse Attention” with Multi-head Latent Attention.

Each of its 77 mixture-of-experts layers has 256 routed experts and one shared expert, with 8 routed experts active per token, according to the card.

Tencent’s own vLLM example serves the FP8 model with a tensor-parallel size of 8, which means splitting it across eight GPUs.

Total and active parameters from the model cards; Hy4 activates about 6.4% of its parameters per token (our ar
Total and active parameters from the model cards; Hy4 activates about 6.4% of its parameters per token (our arithmetic).

The Number

Only about 6.4% of Hy4’s parameters work on any single token: 49 billion of 770 billion (our arithmetic on the card).

DeepSeek’s V4-Pro card lists 1.6 trillion total parameters and the same 49 billion active. Hy4 is about 48% of that size on total parameters (our arithmetic).

The smaller Hy4 flash, which we covered on 16 November, lists 155.5 billion total and 11.3 billion active parameters.

The Licence Terms

Section 10 of the licence defines “MaaS” as giving third parties remote access to model inference or fine-tuning, for example through an API.

A MaaS business needs a separate written licence from Tencent if its worldwide gross revenue exceeds US$50 million in any 12-month period, whether or not that revenue comes from Hy4.

It also needs one before using Hy4 in a product with more than 100 million monthly active users. Purely internal use is exempt if neither the model nor its outputs reach third parties.

What Tencent Claims

Tencent’s pre-training table puts Hy4 base at 91.6 on MMLU against 91.1 for DeepSeek-V4-Pro-Base, and 57.2 on SimpleQA against 52.1. These are Tencent’s own scores.

In a finance test with 213 finance professionals across 579 tasks, Tencent says Hy4 averaged 3.04 out of 4, ahead of DeepSeek-V4.1-Flash-1017 at 2.97 and Kimi K3.1 Pro at 2.68.

Bar chart: Tencent reports Hy4 averaged 3.04 of 4 in its finance test versus 2.97 for DeepSeek-V4.1-Flash-1017 and 2.68 for Kimi K3.1 Pro and GLM-5.3
Tencent’s own finance evaluation from the Hy4 model card; we have not tested the models.

The card also states a limit: “a notable gap remains between it and top proprietary models”, particularly in research-level mathematics and long-horizon coding.

Why It Matters Now

On 8 November we reported that Tencent’s chief AI scientist would not say whether Hunyuan stays open if it takes the lead. The weights of its largest model are now public.

The release lands as US rules tighten Chinese access to advanced chips. Tencent was among the companies that applied for US licences to buy Nvidia H200s, and this week’s rule now covers overseas compute hubs.

The Structural Read

As we read the licence, the revenue trigger counts all of a reseller’s revenue, not just Hy4 revenue, so most large cloud providers would need a deal before offering it.

As we read the card, Tencent is competing on active size: 49 billion active parameters, the same as DeepSeek’s V4-Pro, at about half the total.

Tencent Hy4 model card

“a notable gap remains between it and top proprietary models”

Three Implications

FOR DEVELOPERS The weights are downloadable on Hugging Face, ModelScope and GitCode, with an FP8 version Tencent serves across eight GPUs.

FOR CLOUD RESELLERS Above US$50 million in revenue or 100 million monthly users, a separate Tencent licence is required.

FOR THE OPEN-MODEL RACE Hy4 lists 49 billion active parameters, the same as DeepSeek-V4-Pro, at about 48% of its total size (our arithmetic).

What Is Not Established

The benchmark figures are Tencent’s own, and we have not run the model. Tencent’s card does not give a release date; we date the weights by the Hugging Face listing.

This Week In The AI Supercycle

AI Lost Its Central Banker, the Lab Oligopoly, and the Future of Open Source

This week’s issue: the Nvidia leadership change, read as the AI build-out losing its central banker: who now finances the GPU landlords, backstops the data centres and sponsors open models, and why two labs are becoming the bank.

Read the full issue →

Business Engineer Framework

The AI Supercycle

The Business Engineer map of who controls the bottlenecks in the AI stack.

Read the Map of AI →

The Bottom Line

Tencent has made the 770-billion-parameter Hy4 downloadable, with 49 billion parameters active per token. Its licence keeps a hand on large resellers: above $50 million in revenue or 100 million monthly users, a separate deal with Tencent is required.

Important: not investment advice

This article is for information and education only. It is not investment, financial, legal or tax advice, and it is not a recommendation or an offer to buy or sell any security, ETF, option or other financial product. FourWeekMBA is not a broker-dealer or a registered investment adviser. IPO shares, leveraged and inverse ETFs, and options carry significant risk, including the loss of your entire investment; leveraged and inverse ETFs reset daily and can lose value quickly. Prices, availability, fees and eligibility change and can differ by country and broker, so check the prospectus and your broker’s terms. Figures marked “our arithmetic” are our own calculations from the cited filings and may contain errors. Do your own research and consider speaking to a licensed financial professional before making any investment decision.

97,000+ executives read Business Engineer for the AI strategy frameworks cited by ChatGPT, Claude, and Perplexity.

A note on sourcing. Parameter counts, architecture, context length, deployment example, benchmark scores and the known-limitations line are from Tencent’s Hy4 model card; the commercial terms are from the Tencent Hy License. DeepSeek-V4-Pro figures are from DeepSeek’s model card. Benchmarks are Tencent-reported. The 6.4% active share and the 48% size comparison are our arithmetic.

Sources: Tencent Hy4 model card, Hugging Face (tencent/Hy4), retrieved 20 Nov 2026 · Tencent Hy License, section 10 Additional Commercial Terms · Tencent Hy4 flash model card, Hugging Face · DeepSeek-V4-Pro model card, Hugging Face · Tencent-Hunyuan/Hy4 GitHub repository

Scroll to Top
FourWeekMBA