The Den Reviewthe briefing for executives who run AI
Wednesday, November 11, 2026

The Denominator

Subscribe
Trending

Model Labs · Model Launches

Tencent's Flagship Hy4 Learns to Use Computers, at $0.83 per Million Input Tokens

It's Tencent's first model with computer use, trained on software like Blender and CAD. Weights arrive shortly.

· 2 min read

A man works at a desk with three monitors, showing only abstract shapes, in a high-rise office at dusk with towers beyond the windows.
Photograph: AI-generated

Tencent priced its new 770-billion-parameter Hy4 at the same level as the August preview, a rate a published report put at $0.834 per million input tokens, though the model adds vision and the computer-use skills the preview lacked.

Tencent held the rate while adding capability, but a finance team still cannot price a finished desktop task. Tencent did not say how many tokens Hy4 spends working through a screen, and the open weights that would let a company run the model on its own hardware are still to come.

What does Hy4 cost?

The standard output rate is $2.501 per million tokens, the report said, and Tencent’s own release gives about $0.90 for input and $2.70 for output. On Tencent Cloud’s TokenHub platform, Hy4 is billed at half the standard price through Dec. 31.

OpenRouter lists Hy4 at $0.417 for input and $1.251 for output, half the rates in the report. Two platforms, two prices, and Tencent did not say how long either lower rate lasts beyond the TokenHub date.

What can it do?

Hy4 works through command-line environments and graphical interfaces, with specialized training on professional software such as Blender and CAD. In Tencent’s evaluation inside WorkBuddy, its office assistant, Hy4 earned an average quality score of 83.18 and a pass rate of 96.15%. Tencent published no score for computer use itself.

Vals AI, an outside evaluator, scored Hy4 at 39.4% on Terminal-Bench 4.0, a measure separate from Tencent’s own tests.

What comes next?

Tencent plans to release the weights and a system card “in the coming days,” with no date given. A second model, Hy4 Flash, will keep Hy4’s capabilities at a lower cost per task, Tencent said, and it also set no date.

The price per token is public; the number of tokens a finished task takes is not. The half-price rate on Tencent’s cloud runs only through Dec. 31.

NVIDIAMicrosoftOpenAIAnthropicGoogleAWSJPMorganCapital OneAT&TCharterComcastT‑MobileUnitedHealthCVS HealthTravelersWalmartSalesforceDatabricksSnowflakeCoreWeaveStripeAccenture