Tencent Hy4: $0.83 per Million Tokens for Vision and Computer Use
In short
- Tencent launched Hy4 on November 4: a 770B-parameter model that reads charts and documents and operates command-line tools and graphical interfaces.
- Price: $0.834 per million input tokens and $2.501 per million output tokens. Tencent Cloud TokenHub takes 50% off until December 31.
- The money angle: a batch of 1,000 documents costs about €3.5, and there is a free quota in WorkBuddy and CodeBuddy until November 15.
A big Chinese lab just priced a vision and computer-use model very low. If you process documents or screenshots, this one is worth ten minutes of testing.
What happened
Tencent released Hy4 on November 4. It is a 770-billion-parameter flagship that reads charts and documents, reasons across several images, and can drive command-line tools and graphical interfaces. Tencent trained it on multi-program tasks, including Blender and CAD software.
Pricing is $0.834 per million input tokens and $2.501 per million output tokens, the same as the Hy4 preview. Through Tencent Cloud TokenHub there is a 50% discount until December 31. WorkBuddy and CodeBuddy have a limited free quota from November 4 to November 15. You can also reach it through third-party platforms including OpenRouter. Tencent says the weights and a system card come soon, and a cheaper Hy4 flash model is planned.
How to use it
Use it where you pay people to look at paper. Invoices, price lists, supplier PDFs, screenshots of dashboards, charts from reports. Send the image, ask for a table back.
Do the math: say you process 1,000 supplier documents. Each uses about 3,000 input tokens and 500 output tokens. That is 3 million input tokens (about $2.50) and 0.5 million output tokens (about $1.25), so roughly $3.75 total, or around €3.5. With the TokenHub discount it is close to €1.75. A person typing those 1,000 documents at 3 minutes each spends 50 hours. At €8 an hour that is €400.
Steps: take 20 of your real documents, run them through OpenRouter or the free quota, and compare the output against your own data. Check numbers, not style. If errors stay under what you accept, scale it.
What it means
Chinese labs keep pushing price down while US labs defend premium plans. Hy4 is not proven yet: the benchmarks are Tencent’s own, and the weights are not out, so you cannot run it yourself today. Data also goes through Tencent Cloud or a third-party router, which matters if your documents are sensitive.
Winners are small teams with lots of document work. Losers are tools that charge per page for OCR and extraction.
Ivan’s Take
Worth a test, not a full switch. I would run 20 real documents through the free quota this week and keep customer data out of it until the weights are released. At about €3.5 per thousand documents, even a half-good result pays for itself.
Source: TechNode
If it doesn’t save you time or money, you won’t see it here
- Facts, not hypeEvery price and number is checked against the official source
- Always the real priceYou see what it costs before you click
- I say when it’s weakEven when it’s OpenAI or Google
