Tencent
Hy4
Total Context
256K
Max Output
128K
Released
N/A
hy4-flashhy4-flash has roughly one-quarter the parameters of the flagship model. With lower pricing, it matches flagship open-source performance on everyday tasks — well suited to coding assistance, agent workflows, document and chart understanding, and high-volume daily office work.
Context Window
1M tokens
Maximum Output
65.5K tokens
Release Date
Nov 16, 2026
Modalities
| Input Price | Output Price | Cache Read |
|---|---|---|
| $0.035/M | $0.1405/M | $0.0015/M |
Reasoning
Tool calling
Temperature parameter
Attachments
Knowledge Base
Endpoint Protocols
Hy4 flash pairs low inference cost with visual understanding and long context for recurring work.
The mixture-of-experts architecture activates about 12 billion parameters per token, supporting cost-efficient inference at scale.
Reads visual evidence such as chart labels, diagrams, and screenshots alongside written instructions.
A context window of about one million tokens leaves room to compare long source documents and track earlier decisions.
Handles recurring workflows that gather information, apply it to a draft, and check the result.
Use Hy4 flash for common document, presentation, chart, and coding tasks.
Apply review notes to a draft, preserve its structure, and summarize changes for the next reviewer.
Turn a brief and supporting notes into a slide outline with one takeaway and supporting facts per slide.
Extract labels, units, and visible values from chart images into a reviewable table, flagging any ambiguous entries.
Implement a small change, update its related tests, and summarize the change for review.
import OpenAI from "openai"
const client = new OpenAI({
apiKey: process.env.TOKENHUB_API_KEY,
baseURL: "https://us-api.tokenhub.com/v1",
})
const result = await client.responses.create({})
console.log(result.output_text)Context limits, supported inputs, downloadable weights, pricing, and API access for Hy4 flash.
Hy4 flash is Tencent’s cost-efficient Hy4 model for everyday productivity. It accepts text and images and is designed for routine coding, document work, and workflows that use tools.
Tencent’s model card lists a context window of 1,048,576 tokens. TokenHub shows about 65.5K tokens as the maximum output. Check the selected endpoint for other request limits.
Hy4 flash takes text and images as input and returns text. Use an image-capable endpoint for screenshots or chart images.
Yes. Tencent publishes the instruction-tuned model, an FP8 version, and a pre-trained base model under the Tencent Hy License. See the model card for download and deployment details.
Use your TokenHub API key and set the model field to hy4-flash. The API Integration section provides request examples for Chat Completions, Responses, and Messages.
Tencent Cloud’s published list rates per million tokens are CNY 0.50 for input, CNY 2.00 for output, and CNY 0.02 for cache hits. This TokenHub page’s Pricing table shows the rates for requests sent through this service.
Choose Hy4 flash for frequent office work, document and chart review, or small code changes where cost and turnaround matter. Tencent recommends the flagship Hy4 for more demanding long-running software engineering.
Use one API key to access Hy4 flash and more AI models through TokenHub.
Hy4 Flash In The Community
Launch posts, hands-on demos, and developer discussions about Hy4 flash.
X (Twitter)
Tencent Hy ·
⚡Meet Hy4 flash, the cheapest model for everyday productivity. Matches Claude Haiku 5.5 (high) on daily work (human blind tests), yet 18x task cost reduction. API pricing (per 1M tokens): $0.003 cached · $0.070 input · $0.281 output. 🔗:https://hy.tencent.ai/research/hy4-flash https://t.co/oKjoAQ0ii5
https://pbs.twimg.com/media/HXOhuSrbYAAeHbJ.pngView originalTencent Hy ·
Hy4 flash is now open-weight, including Hy4 flash-FP8 and Hy4 flash-Base. 224B (12B active), 1M context, natively multimodal. Fast, efficient and built for everyday productivity. It's yours now. And stay tuned for Hy4 coming up next! 🤗: https://huggingface.co/tencent/Hy4-flash 🐙: https://t.co/DDckNapEbU
https://pbs.twimg.com/media/HXYXE8aa8AEqG9G.jpgView originalTencent Hy ·
Hy4 flash isn’t just fast and budget-friendly. In real productivity scenarios, it’s already generating a lot of surprisingly useful cases — from daily office work to more complex agent-style tasks. Try it yourself and create your own case. Tag us @TencentHunyuan when you post
View originalReddit
Introducing Hy4 flash
noctrex ·
View originaltencent/Hy4-flash · Hugging Face
coder543 ·
View originalThoughts on hy4 flash so far?
mozkohor ·
Anyone tried it on actual proper work yet? How is it compared to DS v4.1 flash and glm 5.3 flash?
View originalYouTube
Tencent Hy4 Flash Tested: Is a Cheap Flash Model Enough for Daily Work?
Fahd Mirza
View originalTencent Hy4 Flash Is Out. Time to Ditch Qwen 3.8 Flash Next?
RepoChad
View originalI Tested 3 NEW Chinese LLMs: Qwen4-Max, Hy4 Flash, Ernie 5.5 Flash
AI Coding Daily
View original