SHENZHEN, November 4 — Tencent released Hy4 on Wednesday, November 04, introducing what it describes as visual understanding and computer-use capabilities. For small-business owners, availability isn’t evidence of usefulness in customer-facing work.

The model is available through Tencent Cloud TokenHub and workplace products including WorkBuddy, while its underlying model weights remain scheduled for release. Tencent said Hy4 can process images, analyse complex charts and interact with command-line and graphical interfaces, with specialised training on Blender and CAD software. Its API is being made available through third-party platforms including OpenRouter, Hermes and OpenCode.

Tencent’s announcement lists unchanged standard pricing of approximately $0.90 per million input tokens, $2.70 per million output tokens and $0.045 per million cached input tokens. TokenHub is offering a 50% discount through December 31, 2026; WorkBuddy and CodeBuddy are available free from November 4 to 15.

Through WorkBuddy’s integration with WeCom, Hy4 can draw on chats, emails, documents, calendars and meetings to track projects and identify outstanding tasks, Tencent said. Those are company-described capabilities, not evidence of improved sales or customer acquisition. Tencent’s internal evaluations underline the need for review: its reported hallucination rate fell to 7.2% from 12.95% in Hy4 Preview, while errors attributed to unsupported assumptions fell to 5.6% from 23.1%. Those results do not establish accuracy for customer communications. Bloomberg said it had not independently verified Tencent’s benchmark results.

Tencent said it excluded offensive exploitation materials from training. Its Xuanwu Lab and Keen Security Lab assessed areas including cryptographic reverse engineering, binary deobfuscation and vulnerability exploitation, finding no significant offensive-capability concerns in the areas evaluated. Tencent plans to release model weights and system cards in the coming days.

The editorial implication: owners should test a bounded workflow, such as preparing a prospective-customer response, with human review. That’s a test of operational usefulness, not proof that a model generates demand.

Sources