Tencent Hy4 flash post
Architecture, intended uses, list prices and promotional dates.
Model reference · checked 17 Nov 2026
Tencent’s cost-efficient Hy4 model for everyday work: documents, slides, search and routine agents. Below are the published specs, TokenHub list prices, an API quickstart, and the limits you should know before you switch.
Quick answer
Hy4 flash is a language model in Tencent’s Hy4 family, published on 16 November 2026. Tencent’s own post describes it as natively multimodal, with a one-million-token context window, and positions it for everyday work such as deep search, document processing, slide creation, simple frontend demos, and computer use. For long-horizon software engineering, the same post still recommends the larger Hy4 model.
The published size is 224 billion total parameters, including 64 billion Engram parameters, with 12 billion parameters active per token. That is a much smaller active count than the Hy4 flagship. The post says the model combines hybrid attention, LatentMoE, and Engram, and that its KV cache is 1/24 the size of Hy3 and 1/6 the size of Hy4. These are Tencent’s figures, checked against that post on 17 November 2026, not an independent audit.
You can already call it as a hosted API. On Tencent Cloud TokenHub the model id is hy4-flash. The list price on TokenHub’s pricing page is $0.070 per million input tokens, $0.281 per million output tokens, and $0.003 per million cached input tokens. Tencent also wrote that, through 30 November 2026, WorkBuddy users can use Hy4 flash for free and TokenHub API calls are 50% off. Vercel’s changelog, dated 16 November 2026, said the same model is on AI Gateway at 50% off through 29 November 2026. OpenRouter listed tencent/hy4-flash on 17 November 2026, even though Tencent’s post had said it would launch there soon.
Tencent’s post also cites internal evaluations: a WorkIndex task-cost comparison, and a blind test with 85 WorkBuddy users across 259 tasks in 27 categories, where it reported satisfaction close to Claude Haiku 5.5 and a lower cost per task than the other models in that test. Those numbers are vendor claims. We have not reproduced them, and as of 17 November 2026 we did not find a published Artificial Analysis score for Hy4 flash. Read the limitations section before you treat those figures as a buying reason.
Specs
Figures below come from Tencent’s Hy4 flash post or TokenHub API docs. Each row links to its source.
Checked 17 Nov 2026
Quickstart
Activate the model in the TokenHub console first, or the request returns HTTP 402. Replace YOUR_API_KEY with a key from that console. The Singapore global endpoint is shown; Guangzhou and Silicon Valley URLs are on the API usage page.
curl https://tokenhub-intl.tencentcloudmaas.com/v1/chat/completions \
-H "Authorization: Bearer $TOKENHUB_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "hy4-flash",
"messages": [{"role": "user", "content": "Summarize this memo in five bullets."}]
}'import os
from openai import OpenAI
client = OpenAI(
api_key=os.environ["TOKENHUB_API_KEY"],
base_url="https://tokenhub-intl.tencentcloudmaas.com/v1",
)
r = client.chat.completions.create(
model="hy4-flash",
messages=[{"role": "user", "content": "Summarize this memo in five bullets."}],
)
print(r.choices[0].message.content)import OpenAI from "openai";
const client = new OpenAI({
apiKey: process.env.TOKENHUB_API_KEY,
baseURL: "https://tokenhub-intl.tencentcloudmaas.com/v1",
});
const r = await client.chat.completions.create({
model: "hy4-flash",
messages: [{ role: "user", content: "Summarize this memo in five bullets." }],
});
console.log(r.choices[0].message.content);Versions
Happened
Tencent published Hy4 flash and listed TokenHub, WorkBuddy, Yuanbao and other Tencent products as access points, with a WorkBuddy free period and TokenHub 50% off through 30 Nov 2026.
Tencent postHappened
Vercel wrote that tencent/hy4-flash is on AI Gateway with 50% off input, cached input and output through 29 Nov 2026.
Vercel changelogNow
OpenRouter listed tencent/hy4-flash. A Hugging Face search still did not show a Tencent Hy4 flash checkpoint.
Variants
| Feature | Hy4 flash | Hy4 |
|---|---|---|
| Role in Tencent’s post | Everyday productivity | Harder, long-horizon work such as software engineering |
| Total / active parameters | 224B total, 12B active | Larger flagship; flash is described as a leaner variant |
| Context (TokenHub) | 1M, 960k in / 64k out | Same 1M, 960k in / 64k out |
| TokenHub list price per 1M | $0.070 in / $0.281 out / $0.003 cache | $0.834 in / $2.501 out / $0.042 cache, per Tencent’s post |
Source links
Architecture, intended uses, list prices and promotional dates.
Model ids, context limits and protocol notes for hy4-flash.
List rates for Hy4 flash and other TokenHub models.
Base URLs, protocols and the 402 activation rule.
Third-party listing for tencent/hy4-flash, including a promotional rate when shown.
List vs promotional TokenHub rates, with a token-cost estimator.
Nothing matches. Try another keyword or category.
FAQ
On Tencent Cloud TokenHub the list price, checked 17 Nov 2026, is $0.070 per 1M input tokens, $0.281 per 1M output tokens, and $0.003 per 1M cached input tokens. Tencent’s 16 Nov 2026 post said TokenHub API access is 50% off through 30 Nov 2026, and WorkBuddy access is free through the same date. Confirm the live rate on the TokenHub pricing page before you budget.
Not as a Tencent release, as of 17 Nov 2026. The Hy4 flash post says Tencent plans to release Hy4 flash as an open-weight model, but a Hugging Face model search on that date did not show a Tencent Hy4 flash checkpoint. A community repo named Hy4-preview-FlashMoE is a preview-era quant, not the 16 Nov Hy4 flash release. Use TokenHub, WorkBuddy, OpenRouter, or another hosted API until weights are published.
It depends on cache hits and whether DeepSeek is in peak hours. TokenHub list for Hy4 flash is a flat $0.070 input / $0.281 output / $0.003 cache. DeepSeek’s own pricing page bills deepseek-flash (DeepSeek-V4.1-Flash-1017) with peak and off-peak rates: cache-miss input $0.15 off-peak / $0.30 peak, output $0.60 / $1.20, cache-hit input $0.003 / $0.006. Tencent’s post compared WorkIndex task cost with DeepSeek-V4.1-Flash-1017 at Max-tier peak-hour pricing; that is a vendor benchmark, not an independent score. See the comparison page for the sourced table.
As of 17 Nov 2026 we did not find a published Artificial Analysis model page or Intelligence Index score for Hy4 flash. Artificial Analysis does list Tencent Hy4, the larger flagship, as a separate model. Treat any third-party chart that mixes the two names as unverified until you open the underlying page.
On Tencent Cloud TokenHub the model id is hy4-flash. Use the OpenAI-compatible Chat Completions path /v1/chat/completions, or the Responses or Anthropic Messages paths. The Singapore (global) base URL is https://tokenhub-intl.tencentcloudmaas.com/v1. Authenticate with Authorization: Bearer YOUR_API_KEY. Activate the model service in the TokenHub console first; an unactivated call returns HTTP 402. Deep reasoning defaults to enabled for hy4-flash. OpenRouter lists the same model as tencent/hy4-flash.
No. This is an independent reference. Every figure links to the Tencent, DeepSeek, OpenRouter or Vercel page it came from, with a check date of 17 Nov 2026.
Copy the TokenHub quickstart, activate hy4-flash in the console, and confirm today’s rate on the pricing page.