Tencent Hy4 arrived on November 4, 2026, and it changes what a business AI model can do in a normal workday. For example, the new release adds vision and computer use for the first time in the Hunyuan line. Therefore, teams can now ask one model to read a chart, open a tool, and finish a multi-step task. In addition, Tencent kept the listed API price unchanged from the preview, so early adopters can test the new skills without paying a new premium.
However, the model is not a finished open model yet. Because Tencent says it plans to release the weights and a full system card in the coming days, developers cannot download the full model today. As a result, the smart move is to test it through an API first, learn where it helps, and wait for independent checks before you rebuild a workflow around it.
What Tencent Hy4 Is
Tencent Hy4 is the full release of Tencent Hunyuan’s flagship model. Previously, the Hy4 preview launched on August 28, 2026 as a text-only model built for coding and office work. In contrast, the full release adds image understanding and computer operation, and Tencent positions it for workplace productivity, finance work, and game development. For example, it can read fine details in images, analyze complex charts, and solve visual math problems.
Similarly, the model keeps a very large context window. The preview offered a 1 million token context window, and OpenRouter lists about 1 million tokens for the full model as well. Therefore, teams can feed long documents and large project files into one session. In addition, Tencent reported 770 billion total parameters for the preview family, with 49 billion active parameters per token, although it did not restate a parameter count in the full release announcement.
How Tencent Hy4 Vision Works for Business Documents
Tencent Hy4 vision capabilities matter most when work starts with a picture, not with clean text. For example, a finance team can upload a scanned report or a complex chart and ask clear questions about the numbers it shows. Similarly, an operations team can share a product photo and ask the model to spot a detail that a text prompt would miss. As a result, vision turns the model from a chat tool into a document reader for real business files.
However, vision alone does not guarantee accuracy. Because charts can mislead a model, teams should always check key figures against the source document before they act on an answer. Therefore, use the vision features for a first draft or a summary, and keep a human review step for money, safety, and compliance decisions.
Tencent Hy4 Computer Use: From Commands to Clicks
Tencent Hy4 computer use is the biggest change in this release. For example, the model can work through command line tools and graphical interfaces, so it can move beyond answering questions and start operating software. Similarly, Tencent says it trained the model on professional software such as Blender and computer aided design tools, and on long tasks that cross several apps. As a result, a team can ask it to plan a sequence, open the right tool, and work through the steps in order.
Meanwhile, computer use still needs guardrails. Because an agent that clicks and types can also change files or alter settings, teams should run it in a sandbox with limited permissions first. Therefore, start with read only tasks, then add one safe write action at a time, and log every step the agent takes. Finally, never give a new agent full access to production systems on day one.
Tencent Hy4 Pricing: What Teams Actually Pay
Tencent Hy4 pricing stays simple, and that simplicity helps budget planning. The listed API price is $0.834 per million input tokens and $2.501 per million output tokens, with cached input at $0.042 per million tokens. Therefore, cache repeated context when you can, and keep prompts tight.
In addition, Tencent offers two short term savings. Firstly, Tencent Cloud TokenHub offers a 50 percent discount through December 31, 2026. Secondly, WorkBuddy and CodeBuddy offer free use from November 4 to November 15, 2026, so teams can try the assistants before they pay. However, the free window ends quickly, so set a reminder to review usage before November 15. Meanwhile, OpenRouter lists the model at about $0.42 per million input tokens and $1.25 per million output tokens, which matches the TokenHub discount level.
Tencent Hy4 vs Claude Opus 5.5: An Honest Comparison
A Tencent Hy4 vs Claude Opus 5.5 comparison shows a close race in computer use and a wider gap in terminal coding. For example, on OSWorld 2.1, a computer use benchmark, the model scores 78.9, while Claude Opus 5.5 scores 81.8 in Tencent’s own table. Similarly, it scores ahead of GPT-6.1 Sol at 75.2 on the same test. As a result, teams that need screen operation should take it seriously, even though it does not lead the test.
However, the gap widens on terminal work. On Terminal-Bench 4.0, the model reaches 44.4, while Claude Opus 5.5 reaches 66.4 in the same Tencent run. Therefore, developers should not assume that strong screen skills mean equally strong terminal coding skills. In addition, the model leads on only two benchmarks in its own 34 benchmark table, including APEX-Agents v1.1 at 75.2 against 74.8 for Claude Opus 5.5. Because most of these figures come from Tencent’s own tests, treat them as vendor reported results, not as independent proof.
| Test | Tencent Hy4 | Claude Opus 5.5 |
|---|---|---|
| OSWorld 2.1 computer use | 78.9 | 81.8 |
| Terminal-Bench 4.0 | 44.4 | 66.4 |
| APEX-Agents v1.1 | 75.2 | 74.8 |
For context, speed changes value as much as accuracy. Mercury 3 showed how a new design can deliver much faster output. Similarly, our Perplexity vs Claude guide explains why one score never tells the full story. Finally, the NVIDIA Nemotron 3.5 guide shows a local model path when data must stay in house.
How to Access It Through OpenRouter and Tencent Cloud
Tencent Hy4 API OpenRouter access gives teams outside China a simple starting point. For example, developers can call the model through OpenRouter with an existing API setup, test vision and computer use, and compare cost against other models on the same platform. Similarly, Tencent Cloud TokenHub offers direct access, plus the 50 percent discount through December 31, 2026. As a result, many teams will test on OpenRouter first, then move high volume work to TokenHub if the discount makes the math work.
Meanwhile, WorkBuddy, CodeBuddy, Yuanbao, and ima all include the model, and WorkBuddy and CodeBuddy are free from November 4 to November 15. Therefore, a non technical team can start in WorkBuddy and only involve developers when the pilot proves value.
Tencent Hy4 for Business Automation: Where It Fits First
Tencent Hy4 for business automation fits best where work crosses apps, documents, and screens. For example, a finance team can use the model to read reports, check figures, and draft an investment summary, because Tencent reports a finance panel score of 3.04 across 579 real tasks, ahead of DeepSeek-V4.1-Flash at 2.97 in the same blind evaluation. Similarly, a support team can route tickets, gather account details, and draft replies, while a human approves each action.
However, finance and game results still need independent checks. Because Tencent built many of its own test suites, outside teams should rerun a small set of their own tasks before they scale. Therefore, pick five real tasks, score the model on accuracy and time, and compare those results with your current tool.
- Firstly, pick one repetitive workflow
- Secondly, run it in a sandbox with read only access
- Thirdly, score accuracy and cost for at least 20 real cases
- Finally, expand only the steps that pass your bar
Limits You Should Know Before You Switch
Limits matter as much as new features. For example, Tencent itself acknowledges a large gap between the model and the top closed models, so teams should not expect it to beat every frontier model on every task. Similarly, the full weights are not public yet, so you cannot inspect, fine tune, or self host the full model today. As a result, any plan that depends on local deployment must wait, and API access remains the only route for now.
Meanwhile, benchmark honesty cuts both ways, because Tencent’s 34 benchmark table mixes public tests with 20 internal tests. Therefore, read the source notes, not just the score. In addition, Unite.AI coverage confirms the vision and computer use additions and the unchanged listed prices, while StartupFortune coverage adds the free window and the December discount deadline.
Frequently Asked Questions
What is it?
Tencent Hy4 is Tencent Hunyuan’s full flagship model, released on November 4, 2026, with vision and computer use added to the August preview. For example, it can read images and charts and operate command line and graphical interfaces.
How much does pricing cost?
Tencent Hy4 pricing lists $0.834 per million input tokens and $2.501 per million output tokens, with cached input at $0.042 per million tokens. In addition, TokenHub offers a 50 percent discount through December 31, 2026.
Does computer use work on normal apps?
Yes, Tencent Hy4 computer use works through command line tools and graphical interfaces, and Tencent trained it on professional software such as Blender and CAD tools. However, teams should still test it in a sandbox before they give it production access.
How does vision help business teams?
Tencent Hy4 vision capabilities help teams read charts, documents, and photos without manual transcription. For example, a team can compare two charts, extract a figure, and draft a summary for human review.
Can I access the API through OpenRouter?
Yes, Tencent Hy4 API OpenRouter access is live, and OpenRouter lists about 1 million tokens of context with image input. Therefore, developers outside China can test the model without a Tencent Cloud account first.
Is it open source today?
No, the full weights are not public yet. Because Tencent says it plans to release them in the coming days, along with a system card, teams should treat the model as API only until that release actually lands.
Should small businesses try it for business automation?
Yes, for a small pilot, because the free WorkBuddy and CodeBuddy window runs from November 4 to November 15, 2026. Therefore, a small team can test one workflow at no cost, measure time saved, and decide with real data.
Does it beat Claude Opus 5.5?
Not overall, based on Tencent’s own table. For example, the model trails Claude Opus 5.5 on OSWorld 2.1 and Terminal-Bench 4.0, but it edges ahead on APEX-Agents v1.1. As a result, choose the model by task, not by headline.
When will Hy4 Flash arrive?
Tencent says a faster, more cost efficient Hy4 Flash version will follow soon, but it has not announced a date. Meanwhile, pilot the full model now and keep routine tasks ready to move when Flash lands.
What should I do first?
Firstly, try WorkBuddy or CodeBuddy during the free window. Secondly, pick one repetitive workflow, run it in a sandbox, and score accuracy and cost. Finally, expand only the steps that pass your bar.
Conclusion
Tencent Hy4 earns attention because it adds two practical skills at an unchanged price: vision for documents and computer use for software. For example, teams can read charts, operate apps, and draft work faster, while the TokenHub discount through December 31, 2026 lowers the cost of a serious pilot. Therefore, the best move is a small, measured test, not a full migration on day one.
However, honesty still matters. Because the weights are not public and the benchmarks are largely Tencent’s own, teams should verify every claim on their own tasks. As a result, start with the free window, run a sandbox pilot, and let your own accuracy and cost numbers decide.
