Tencent Hy4 API: how to use it on OpenRouter and what it costs

You can use the Tencent Hy4 API today through OpenRouter, with an ordinary OpenRouter key and the model ID tencent/hy4. It is not free. OpenRouter lists $0.417 per million input tokens and $1.251 per million output tokens for it, as of November 4, 2026. The older preview, tencent/hy4-preview, costs twice that, and its weights are public.

Tencent released the official Hy4 on November 4, adding image input and computer use to the text-only August preview. We covered the launch in Tencent releases Hy4 with vision and computer use, weights to follow. This page covers what a developer needs: how to call it, what it costs, and what you can download.

How to call Hy4 through OpenRouter

OpenRouter exposes Hy4 through its standard Chat Completions endpoint. You need an OpenRouter API key, sent as a bearer token. This is the example from OpenRouter's page for the official model:

curl -X POST https://openrouter.ai/api/v1/chat/completions \
  -H "Authorization: Bearer $OPENROUTER_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
  "model": "tencent/hy4",
  "messages": [{"role": "user", "content": "What is the meaning of life?"}]
}'

To use the preview instead, change the model to tencent/hy4-preview. Both pages list stream, tools, tool_choice, response_format, reasoning, reasoning_effort, temperature and max_tokens among the supported parameters. The preview page also lists sampling options such as top_p, top_k, min_p and seed.

OpenRouter says it does not offer EU in-region routing for either model. Errors follow its usual scheme: 402 means you are out of credits, 404 means the model or a provider for it was not found, and 429 means you are rate limited.

Is Hy4 free?

Not through the API. Neither OpenRouter page shows a free variant, and both bill per token.

Tencent's own free offer is separate. According to TechNode, Hy4 is built into Tencent's WorkBuddy and CodeBuddy products with a limited free quota from November 4 to November 15. TechNode also reports a 50% discount on Tencent Cloud's TokenHub through December 31.

Hy4 API price

Prices are from OpenRouter on November 4, 2026. Both pages mention one more rate that the text I read does not show, and OpenRouter says Tencent Cloud bills by UTC time of day, so the listed price may apply to only part of the day.

One discrepancy: TechNode reports the standard Hy4 price as $0.834 input and $2.501 output, the same as the preview. That matches the preview row above, not the official model's. Check the live OpenRouter page before you budget. A 50% TokenHub discount would explain a gap, but no source I read says so.

Are the Hy4 weights available?

The preview's weights are. The model card on Hugging Face lists an Apache 2.0 license, a 770 billion parameter mixture-of-experts model (which runs only part of its network for each token) with 49 billion active per token, and a 1 million token context window. There is a BF16 version and an FP8 version. Tencent calls it an early version and says it can reason longer than needed and over-verify its own work.

Running it yourself is heavy. The card gives vLLM and SGLang examples with 8-way tensor parallelism, and sets reasoning to "high" by default. Passing reasoning_effort as no_think gives direct answers.

For the official Hy4, the weights are not out. TechNode says Tencent will release them, with a system card, "shortly". No date or license has been given. Tencent's research page for Hy4, hy.tencent.ai/research/hy4, did not load for me, so the details here come from OpenRouter, Hugging Face and TechNode.

What we don't know yet

• Whether the official weights will carry the preview's Apache 2.0 license.
• Whether the price on OpenRouter is the one Tencent intends to charge long term.
• How the official model scores on independent benchmarks. None were available on launch day.
• The date and price of the cheaper Hy4 Flash that Tencent says is coming.