Skip to content

AI Tools

What Is Tencent Hy4 Flash?

Quick Answer

Tencent Hy4 Flash is a low-cost AI model from Tencent, launched November 16, 2026. It is a distilled version of the flagship Hy4, with 224 billion total parameters and 12 billion active per token, a 1-million-token context window, and native image understanding, tool calling and structured output. It costs $0.07 per million input tokens and $0.281 per million output tokens on Tencent Cloud TokenHub (Singapore list), with a 50% launch discount until November 30, 2026. It is available through the TokenHub API (model id hy4-flash), the WorkBuddy office app, Yuanbao, WeCom, QQ, ima and Marvis. The weights have not been released yet — Tencent says it plans to open-source the model, with no date given.

What Is Tencent Hy4 Flash?

Every model family eventually gets a cheap tier, and Hy4 Flash is Tencent’s. The flagship Hy4, generally available since November 4, 2026, is a 770-billion-parameter model aimed at long-horizon work such as software engineering and cross-application tasks. Hy4 Flash is built by distilling that flagship and then training it with reinforcement learning to say less — shorter answers, fewer wasted tokens — so the same everyday job costs far less to run.

Tencent positions it for everyday office work rather than the hardest reasoning: deep search, writing and revising documents, building slides, spreadsheet work, frontend coding and computer use. The flagship stays the pick for long-horizon software engineering; Flash is the model for the tasks that repeat all day.

It is a mixture-of-experts model: 224 billion total parameters, of which 64 billion are Engram parameters, with 12 billion active on any single token. That is roughly a quarter of the flagship’s active parameters, which is where most of the price gap comes from — the rest comes from serving changes and the shorter outputs the model is trained to produce.

Release Date and Availability

Tencent launched Hy4 Flash on November 16, 2026. TechNode reported the launch the same day, and Tencent Cloud’s TokenHub documentation and model list were updated on November 16 to carry the model, its API prices and its context limits.

It is available now through:

  • Tencent Cloud TokenHub — API access, model id hy4-flash; supports OpenAI and Anthropic protocols
  • WorkBuddy — Tencent’s office app, free to use until November 30, 2026
  • Yuanbao — Tencent’s consumer AI app
  • WeCom, QQ, ima and Marvis — Tencent’s enterprise and consumer surfaces

OpenRouter listing is coming soon, according to Tencent. The model is hosted-only for now: there are no downloadable weights, no licence and no model card, though Tencent says it plans to release Hy4 Flash as an open-weight model and to update it frequently.

Key Hy4 Flash Features

1. A 1-million-token context window

Hy4 Flash keeps the flagship’s full context: 1,048,576 tokens, with a maximum input of 960,000 tokens and a maximum output of 64,000 tokens. Long documents, big codebases and multi-file research fit in a single call.

2. Native image understanding

The model is natively multimodal. It reads images, charts and diagrams as part of the prompt, which matters for slide work, document review and visual research.

3. Tool and function calling

Hy4 Flash can call tools and functions, so it works inside agents and workflows that fetch data, run commands or chain steps — not just in a chat box.

4. Structured output

It supports structured output, so applications can ask for JSON or another fixed format and rely on getting it back, which is what office automation and data pipelines need.

5. Prompt caching

Cached input costs $0.003 per million tokens on the Singapore list — about 23 times cheaper than fresh input — so repeated system prompts and long shared documents get much cheaper on the second call.

6. Deep reasoning on a budget

Tencent describes it as a deep-reasoning model, and its guidance is to use Flash for deep search, documents, slides and frontend work, keeping the flagship for long-horizon engineering. The reasoning is tuned for everyday tasks rather than the hardest problems.

Hy4 Flash Pricing Explained

Hy4 Flash is priced per million tokens on Tencent Cloud TokenHub. The Singapore list price is $0.07 per million input tokens, $0.281 per million output tokens and $0.003 per million cached input tokens. The mainland China list is ¥0.5 input, ¥2 output and ¥0.02 cached per million tokens.

Item Hy4 Flash price
Input tokens $0.07 per 1M tokens
Output tokens $0.281 per 1M tokens
Cached input tokens $0.003 per 1M tokens
Launch discount 50% off list price until 12:00, November 30, 2026 (Beijing time)
WorkBuddy app Free until November 30, 2026

TokenHub also offers a Standard plan at $17 per month with 2,600 credits per month.

A worked example: a task that uses 1 million input tokens and 500,000 output tokens costs $0.07 + (0.5 × $0.281) = about $0.21 at list price. During the launch discount that halves to about $0.11. The same task on the full Hy4 would cost $0.834 + (0.5 × $2.501) = about $2.08 — roughly ten times more.

Hy4 Flash vs Hy4

Spec Hy4 Flash Hy4 (flagship)
Total parameters 224B (64B Engram) 770B
Active parameters per token 12B 49B
Context window 1M tokens 1M tokens
Input price per 1M tokens $0.07 $0.834
Output price per 1M tokens $0.281 $2.501
Cached input price per 1M tokens $0.003 $0.042
Launch date November 16, 2026 November 4, 2026
Best for Everyday office work, documents, slides, search Long-horizon engineering, cross-app work

The price gap is not a discount: it is roughly four times fewer active parameters plus serving changes and shorter outputs. Hy4 Flash costs about 12 times less per input token and about 9 times less per output token than the flagship.

Performance and Benchmarks

All performance figures here are Tencent’s own, not independently verified. There is no public model card, no architecture paper and no independent benchmark run yet.

Tencent ran a blind test on 259 everyday work tasks with its own employees. Hy4 Flash scored 2.79 out of 4 on user satisfaction, level with Claude Haiku 5.5 at 2.81 and ahead of GLM-5.3-Flash at 2.53. The same test put its cost per task at ¥0.26 against ¥4.73 for Claude Haiku 5.5, with a median task time of 6.7 minutes — the shortest of the four models tested.

Tencent also reports per-task cost on its internal WorkIndex suite — eight benchmarks covering PowerPoint creation and revision, co-writing, Markdown writing and similar office work — at one-thirteenth of Hy4 preview and about one-twentieth of the full Hy4. WorkIndex is Tencent’s own suite, so treat the ratio as the company’s claim, not a measured fact. The benchmark charts were published as images rather than tables, and no independent reproduction exists yet.

How to Use Hy4 Flash

1. Use it free in WorkBuddy

Open Tencent’s WorkBuddy office app and use Hy4 Flash free until November 30, 2026. This is the fastest way to try it with no API setup.

2. Get a TokenHub API key

Create a Tencent Cloud account, open TokenHub, and create an API key. TokenHub supports the OpenAI and Anthropic protocols, so existing code written for those APIs can point at it with a changed base URL.

3. Call the model by its id

Send requests with the model id hy4-flash. A simple test is a short document question with a small input, so you can see the output length and cost before scaling up.

4. Cache repeated prompts

Put instructions and long documents that you reuse into cached input, which costs $0.003 per million tokens instead of $0.07 — the cheapest way to run repeated office workflows.

5. Pick the right tier

Use Hy4 Flash for documents, slides, search, spreadsheets and frontend work. Keep the full Hy4 for long-horizon software engineering and multi-application tasks.

Best Uses

  • Office documents — drafting, revising and summarising long documents and reports
  • Slides — creating and revising PowerPoint decks, including chart and image reading
  • Deep search and research — multi-step research across long sources in the 1M-token window
  • Spreadsheets and data work — structured output for tables, extraction and analysis
  • Frontend coding — short coding tasks, prototypes and UI work
  • Everyday agents — tool-calling workflows where the per-task cost matters more than frontier reasoning

Limitations and Things to Know

  • No weights yet. Tencent says it plans to open-source Hy4 Flash, but there is no date, no licence and no repository. For now it is hosted-only.
  • No model card. There is no public model card, architecture paper or independent benchmark. Every performance number comes from Tencent.
  • The launch discount ends November 30, 2026. The 50% TokenHub discount ends at noon Beijing time on November 30, and WorkBuddy stops being free the same day. The list price is what survives December.
  • Shorter answers by design. The model is trained to talk less, which saves money but means it may give briefer answers than a flagship model on open-ended questions.
  • Not the flagship. For long-horizon software engineering and complex multi-application work, Tencent itself points users to the full Hy4.
  • OpenRouter is not live yet. The model is listed as coming soon on OpenRouter; on launch day the only API route is Tencent’s own TokenHub.

Frequently Asked Questions

What is Tencent Hy4 Flash?

Hy4 Flash is Tencent’s low-cost AI model, launched November 16, 2026. It is a distilled version of the flagship Hy4 with 224 billion total parameters, 12 billion active per token, a 1-million-token context window, and pricing of $0.07 per million input tokens and $0.281 per million output tokens.

How much does Hy4 Flash cost?

On Tencent Cloud TokenHub (Singapore list) it costs $0.07 per million input tokens, $0.281 per million output tokens and $0.003 per million cached input tokens. A 50% launch discount runs until November 30, 2026, and it is free in the WorkBuddy app until the same date.

How is Hy4 Flash different from Hy4?

Hy4 is the flagship: 770 billion total parameters, 49 billion active, $0.834/$2.501 per million input/output tokens, built for long-horizon engineering. Hy4 Flash is the cheap tier: 224 billion total, 12 billion active, $0.07/$0.281, built for everyday office work. Both share the 1-million-token context window.

What is the context window of Hy4 Flash?

1,048,576 tokens, with a maximum input of 960,000 tokens and a maximum output of 64,000 tokens per call.

Where can I use Hy4 Flash?

Through the TokenHub API (model id hy4-flash), the WorkBuddy office app, Yuanbao, WeCom, QQ, ima and Marvis. An OpenRouter listing is coming soon.

Is Hy4 Flash free?

It is free in Tencent’s WorkBuddy app until November 30, 2026, and TokenHub is running a 50% discount until noon Beijing time on November 30, 2026. After that, the list price applies.

Is Hy4 Flash open source?

Not yet. Tencent says it plans to release the weights as an open-weight model, but no date, licence or repository has been announced.

What is the model id for the Hy4 Flash API?

hy4-flash. TokenHub supports OpenAI and Anthropic protocols, so you can call it with the model id hy4-flash from existing API code.

Final Verdict

Hy4 Flash is the most interesting kind of launch: not a capability jump, but a price cut that changes what the model is for. Keeping the flagship’s 1-million-token context while charging about one-twelfth as much per input token makes it a genuine everyday-work model rather than a demo. The 50% launch discount and free WorkBuddy access until November 30, 2026 make it cheap to try right now. The honest limits are that every performance claim is Tencent’s own, there is no model card and no weights yet, and the list price — not the discount — is what you will pay in December. If your work is documents, slides, search and short coding tasks, Hy4 Flash is worth trying this month; if it is long-horizon engineering, the full Hy4 is still the pick.

Editorial update: November 17, 2026. This article is based on Tencent Cloud’s TokenHub documentation and pricing, TechNode’s launch report and secondary coverage of the November 16, 2026 launch. Performance figures are Tencent’s own and have not been independently verified.