Unofficial guide
Hy4 Flash API
Last updated
Hy4 Flash is a hosted model on Tencent Cloud TokenHub, not only a weight file. The model id is
hy4-flash. Prices below are copied from Tencent's pricing pages and were current
as of those pages' update dates.
Where to call it
The API usage page gives
three default base URLs. Authentication is Authorization: Bearer and an API key
from the TokenHub console.
| Site | Base URL | Scheduling scope |
|---|---|---|
| Singapore | https://tokenhub-intl.tencentcloudmaas.com | Global |
| Guangzhou | https://tokenhub.tencentcloudmaas.com | Chinese mainland |
| Silicon Valley | https://tokenhub-us.tencentcloudmaas.com | United States |
The Hy API Guide says
hy4-flash and hy4 speak the OpenAI Chat Completions API, OpenAI
Responses, and Anthropic Messages. Its chat example posts to
/v1/chat/completions on the Singapore host. The guide's sample body uses
hy4. Swap that field to hy4-flash to call this model. Temperature in
the sample is 0.9, the same value the weight card recommends.
curl -X POST 'https://tokenhub-intl.tencentcloudmaas.com/v1/chat/completions' \
-H 'Content-Type: application/json' \
-H 'Authorization: Bearer YOUR_API_KEY' \
-d '{
"model": "hy4-flash",
"messages": [
{"role": "user", "content": "Hello, please briefly introduce yourself."}
],
"stream": false,
"temperature": 0.9
}'
The guide's capability row for hy4-flash is a 1M context window, 960k maximum
input, and 64k maximum output. It describes the model as the leaner everyday model for
high-volume work.
List price
On the English pricing page
updated 19 November 2026, hy4-flash is listed at 0.07 USD per million input tokens,
0.281 USD per million output tokens, and 0.003 USD per million cache tokens, with no separate
peak and off-peak rate in that row. The same table lists hy4 at 0.834, 2.501, and
0.042 USD. Tencent's 16 November announcement uses the same flash figures: about 0.070, 0.281,
and 0.003 US dollars per million tokens.
That pricing page says Hy4 Flash inference is billed at 50 percent of list price until 12:00:00 on 30 November 2026, Beijing time, and that Hy4's own 50 percent discount runs until 23:59:59 on 31 December 2026, Beijing time. After the window, fees return to list price. The mainland Chinese pricing page, updated 16 November 2026, lists Hy4 Flash at 0.5 CNY input, 2 CNY output, and 0.02 CNY cache per million tokens, with the same 30 November noon cutoff. This guide does not convert between the two currencies.
Products named at launch
The 16 November announcement says Hy4 Flash is available across Tencent Cloud TokenHub, WorkBuddy, Yuanbao, WeCom, QQ, ima, and Marvis, and that OpenRouter integration was coming soon. It also says WorkBuddy access was free through 30 November 2026 and TokenHub was 50 percent off standard pricing for that period. Those are promotional dates. Confirm them in the product, because this page is not the billing system.
An OpenRouter model page for tencent/hy4-flash existed when this guide was checked.
Live latency and the price a router displays move. Use Tencent's pricing page for the list
price, and the router for whatever it charges today.
Running the weights yourself is a different bill: GPU time, not tokens. That path is vLLM, SGLang, or a rented GPU.
Common questions
Is there an official Hy4 Flash API?
Yes. Tencent Cloud TokenHub documents hy4-flash as an OpenAI-compatible chat model, and also as compatible with the OpenAI Responses and Anthropic Messages protocols. You need a Tencent Cloud account, TokenHub activated, and an API key from the TokenHub console.
What does the Hy4 Flash API cost?
The English TokenHub pricing page, updated 19 November 2026, lists hy4-flash at 0.07 USD per million input tokens, 0.281 USD per million output tokens, and 0.003 USD per million cache tokens. The same page says inference fees are 50 percent of list price until 12:00:00 on 30 November 2026, Beijing time. The mainland Chinese pricing page lists 0.5, 2, and 0.02 CNY per million tokens for input, output, and cache. Re-check both pages. The discount window is short.
What is the context limit on the API?
The Hy API Guide lists hy4-flash with a 1M context window, 960k maximum input, and 64k maximum output. The same row gives hy4 and hy4-preview those same three limits.
Sources
- Tencent Cloud, Hy API Guide TokenHub model ids, context, and a chat-completions example. Page updated 16 November 2026.
- Tencent Cloud, TokenHub API usage Regional base URLs.
- Tencent Cloud, TokenHub model pricing English list prices. Page updated 19 November 2026.
- Tencent Cloud (China), TokenHub model pricing Mainland list prices in CNY. Page updated 16 November 2026.
- Tencent, “Tencent Launches Hy4 flash” Announcement dated 16 November 2026.