Skip to content
Neural Daily

ModelsNews

Tencent Hy4 adds vision and computer use, weights later

🇭🇺 Olvasd el magyarul →

Tencent released the final Hy4: its first Hy model with vision and computer use. API pricing stays the same, and open weights are promised for later.

By: Mikola Máté Patrik · · 2 min read

Tencent Hy4 benchmark chart comparing Hy4 with GLM 5.3, DeepSeek V4.1 Flash, Kimi K3.1 Pro, Claude Opus 5.5 and GPT 6.1 Sol
Image: Tencent · Press kit · Official benchmark image from the Hy4 release page · Source

What happened?

On November 4, Tencent released Hy4, the new flagship of its Hunyuan model family (Tencent Hy). It follows the Hy4 preview from August with larger pre-training and post-training, adds vision for the first time and, also for the first time, can operate a computer through both command-line and graphical interfaces.

According to Chinese media, the preview was a 770B-parameter mixture-of-experts model with 49B active parameters and a 1M-token context window. Tencent says Hy4 is the first Hy model to rank among the strongest open models. The weights are not out yet: Tencent plans to release them together with a detailed system card.

What it was trained for

  • Computer use: targeted training on Blender, CAD and long workflows that span several applications.
  • Game development: support for Unreal Engine, Unity and Godot; versus the preview, scores rose 19.1% on GameDev and 9.8% on GameLogic.
  • Finance: 213 professionals rated it on 579 real tasks. Hy4 ranked first among open models (3.04), ahead of DeepSeek-V4.1-Flash (2.97), Kimi K3.1 Pro and GLM-5.3 (both 2.68).
  • Hallucination: in internal tests the rate fell from 15.76% on the preview to 5.45%.

How it compares with closed models

Tencent is upfront that a clear gap remains to the top proprietary models, especially in research-level math, long-horizon coding and advanced visual question answering. In its own benchmark table (the cover image of this article), Hy4 scores 44.4 on Terminal Bench 4.0 versus 66.4 for Claude Opus 5.5 and 66.8 for Kimi K3.1 Pro. On OSWorld 2.1 it is closer: 78.9 versus 81.8 for Claude Opus 5.5.

Pricing and availability

Hy4 is available on Tencent Cloud and OpenRouter and inside CodeBuddy, WorkBuddy, Yuanbao and ima. API pricing is unchanged from the preview: $0.834 per million input tokens, $2.501 per million output tokens and $0.042 for cached input. A cheaper, faster Hy4 flash is planned soon.

Why it matters

The Chinese open model field (DeepSeek, Kimi, GLM, Qwen) keeps getting more crowded, and Tencent now competes on both price and agentic skills. For teams that want to self-host a large model, the actual weight release is what counts; until then Hy4 is API-only.

    Produced with the help of AI, reviewed by an editor. How we make our articles

    Reviewed and edited by the Neural Daily desk. About

    Related articles