How much does Tencent Hy4 preview cost to use through an API?

Tencent Hy4 preview costs $0.834 per million input tokens and $2.501 per million output tokens, with cache hits billed at $0.042 per million. Those rates apply through Tencent Cloud TokenHub and OpenRouter. The weights themselves are free to download under Apache-2.0, so self-hosting cost is whatever your own hardware bill comes to.

Is there a free way to try Hy4 preview?

Yes. Tencent made Hy4 preview free on WorkBuddy and CodeBuddy for two weeks from launch, and extended free access to the older Hy3 model through September 30. Both routes avoid any API bill. Downloading the Apache-2.0 weights from Hugging Face, ModelScope, GitCode or CNB is also free, though running 770B parameters needs serious hardware.

How does Hy4 preview compare to GLM-5.3 and Kimi K3?

Tencent ran a blind evaluation in which 163 internal experts rated model outputs on 203 engineering tasks. Hy4 preview averaged 2.99 out of 4.00, slightly ahead of Kimi K3 at 2.94 and GLM-5.3 at 2.92. The margin is small, and the test is Tencent's own, so treat it as a close three-way result rather than a clear win.

What are the known weaknesses of Hy4 preview?

Tencent calls Hy4 preview an early version with real headroom left in both pre-training and post-training, and names two shipping issues: the model spends longer than necessary reasoning through complex tasks, and it tends to over-verify its own work. Both cost extra tokens and latency, and Tencent says it plans to address them in later versions.

Where can I download the Hy4 preview weights?

Hy4 preview weights are published on Hugging Face at tencent/Hy4-preview, and mirrored on ModelScope, GitCode and CNB. Tencent offers both a standard and an FP8-quantized version, and the inference code sits in the Tencent-Hunyuan/Hy4-preview repository on GitHub. Everything is covered by the Apache-2.0 license.

Tencent Hunyuan · 2026-08-28 · major

Tencent Hy4 preview — 770B open-weights model with a 1M-token context

Tencent Hy4 preview is a 770B open-weights mixture-of-experts model with 49B active parameters, a 1M-token context and an Apache-2.0 license. In Tencent's blind test, 163 internal experts scored it slightly ahead of GLM-5.3 and Kimi K3.

Cover graphic for Tencent's Hy4 preview open-source model announcement

Tencent open-sources Hy4 preview, a 770B mixture-of-experts model with 49B active parameters and a 1M-token context.

Quick facts

Maker	Tencent Hunyuan
Parameters	770B total, 49B active
Architecture	MoE — 78 layers, 256 routed experts
Context window	1M tokens
License	Apache-2.0
Price (input)	$0.834 / 1M tokens
Price (output)	$2.501 / 1M tokens

Benchmarks

Tencent blind expert rating (203 engineering tasks)

Hy4 preview		2.99 / 4.00
Kimi K3		2.94 / 4.00
GLM-5.3		2.92 / 4.00

source ↗

Pricing

Input	$0.834 / 1M tokens
Output	$2.501 / 1M tokens
Cache hit	$0.042 / 1M tokens

source ↗

What is it?

Hy4 preview opens the weights of Tencent's largest model so far — 770 billion total parameters, 49 billion active per token — under the Apache-2.0 license. Tencent Hunyuan aimed it at productivity work: coding, office tasks, data analysis and scientific research. It went live on the same day inside WorkBuddy, CodeBuddy, Yuanbao and ima, and through the Tencent Cloud TokenHub and OpenRouter APIs.

How does it work?

A mixture-of-experts layout does the heavy lifting: 78 layers, 256 routed experts, and only 49B of the 770B parameters active on any token, which keeps serving cost far below a dense model of the same size. Context runs to 1M tokens. Tencent ships the model with known issues named in the card — it spends longer than needed reasoning through hard tasks, and it tends to over-verify its own work.

Why does it matter?

Open weights at this size give teams a frontier-scale model they can host themselves, and Apache-2.0 allows commercial use without a separate agreement. Teams that do not want to run 770B parameters can rent it instead at $0.834 per million input tokens. Tencent backs the release with a blind test in which 163 internal experts rated outputs on 203 engineering tasks, scoring Hy4 preview 2.99 out of 4.00 against 2.94 for Kimi K3 and 2.92 for GLM-5.3.

Who is it for?

teams self-hosting large open models, agent and coding-tool builders

Frequently asked questions

How much does Tencent Hy4 preview cost to use through an API?: Tencent Hy4 preview costs $0.834 per million input tokens and $2.501 per million output tokens, with cache hits billed at $0.042 per million. Those rates apply through Tencent Cloud TokenHub and OpenRouter. The weights themselves are free to download under Apache-2.0, so self-hosting cost is whatever your own hardware bill comes to.
Is there a free way to try Hy4 preview?: Yes. Tencent made Hy4 preview free on WorkBuddy and CodeBuddy for two weeks from launch, and extended free access to the older Hy3 model through September 30. Both routes avoid any API bill. Downloading the Apache-2.0 weights from Hugging Face, ModelScope, GitCode or CNB is also free, though running 770B parameters needs serious hardware.
How does Hy4 preview compare to GLM-5.3 and Kimi K3?: Tencent ran a blind evaluation in which 163 internal experts rated model outputs on 203 engineering tasks. Hy4 preview averaged 2.99 out of 4.00, slightly ahead of Kimi K3 at 2.94 and GLM-5.3 at 2.92. The margin is small, and the test is Tencent's own, so treat it as a close three-way result rather than a clear win.
What are the known weaknesses of Hy4 preview?: Tencent calls Hy4 preview an early version with real headroom left in both pre-training and post-training, and names two shipping issues: the model spends longer than necessary reasoning through complex tasks, and it tends to over-verify its own work. Both cost extra tokens and latency, and Tencent says it plans to address them in later versions.
Where can I download the Hy4 preview weights?: Hy4 preview weights are published on Hugging Face at tencent/Hy4-preview, and mirrored on ModelScope, GitCode and CNB. Tencent offers both a standard and an FP8-quantized version, and the inference code sits in the Tencent-Hunyuan/Hy4-preview repository on GitHub. Everything is covered by the Apache-2.0 license.

Try it

Download tencent/Hy4-preview from Hugging Face, or call tencent/hy4-preview on OpenRouter.