Text and chat · Tencent

Hy

Tencent’s Hy4 preview: an open 770B mixture-of-experts model with 1M context, tuned for code, office work and agents.

Private Moderate cost
Latest versionHy4 previewAugust 2026
Price$1.406 per million tokens written$0.000984 a page written
Key spec1M tokens of context

What sets it apart

Hy is the model line of Tencent’s Hy team. Hy4 preview is a large open-weight model, text only, which Tencent itself calls an early version with known issues.

  • 770B parameters, 49B active per token
  • Apache 2.0 open weights
  • 1M-token context, text in and out

Strengths and limits

Where it shines

  • Tencent trained it on work from its own engineers, game developers and analysts: software, office documents, data analysis and game prototypes.
  • Weights are free to use and modify under Apache 2.0, with an FP8 version also published.

Think twice when

  • Text only: it does not read images, audio or video.
  • Tencent lists known issues: it can reason longer than needed on complex tasks and over-verify its own work.
  • Hy4 preview was published on Hugging Face in late August 2026.
  • It writes up to 64,000 tokens per reply in this catalog and supports a reasoning effort of none, low or high.

Use cases

Turn files into a shareable report

Here are three spreadsheets and two memos. Build a one-page summary of the quarter with the main variances explained.

Tencent trained it on office and data analysis work.

Prototype a small game

Write a playable browser version of a falling-blocks puzzle game with a score and a game-over screen.

Game prototyping is one of the four areas Tencent names for this model.

Against its rivals

  • GLMIn Tencent’s own blind test of 203 engineering tasks it rated Hy4 preview slightly ahead of GLM 5.3 (2.99 vs 2.92), a vendor claim.#13 of 90 on Artificial Analysis Intelligence Index (45 points)
  • KimiTencent’s own blind test rated Hy4 preview slightly ahead of Kimi K3 (2.99 vs 2.94), a vendor claim.#15 of 90 on Artificial Analysis Intelligence Index (44 points)
  • DeepSeekAnother large open-weight family from a Chinese lab, with 1M context.#24 of 90 on Artificial Analysis Intelligence Index (39 points)
Compare side by side ↗

Version history

Each version, dated, with what it changed over the one before.

  1. Hy4 previewLatest
    • Open-weight 770B mixture-of-experts model with 49B active parameters and a 1M-token context
    • Post-trained for software engineering, office and data analysis, game development and research
    Source : huggingface.co

The model in this family

ModelUsePrivacyPrice
Hy4 Previewhy4-previewTextPrivate$1.406 per million tokens written$0.000984 a page written

Specifications

SpecificationHy4 preview
ReleasedAugust 2026
Parameters770B total, 49B active
WeightsOpen, Apache 2.0
Acceptstext
Producestext
Context1,049K tokens
Longest answer64K tokens
AbilitiesUses Sub Rosa’s tools, Thinks before answering, Tuned for code
Price$1.406 per million tokens written
PrivacyPrivate

Prices, context and lengths from the live catalog on October 16, 2026; parameters and licenses from the maker.

Sources

Everything on this page comes from these pages, read on October 16, 2026 or as dated on each score.

Show the 2 sources
Learn more about Hy ↗