Technical specifications of Tencent Hy4
| Item | Tencent Hy4 |
|---|---|
| Model ID | hy4 (CometAPI listing: hy4) |
| Provider | Tencent |
| Release date | November 4, 2026 |
| Model family | Tencent Hy / Hunyuan |
| Model type | General-purpose large language model focused on agentic and productivity tasks |
| Input modalities | Text and images |
| Output modality | Text |
| Context window | 1,048,576 tokens (1M) |
| Maximum completion | 65,536 tokens |
| Tool/function calling | Supported |
| Structured output | JSON Schema via response_format, according to CometAPI |
What is Tencent Hy4?
Tencent Hy4 is the November 4, 2026 official release in Tencent's Hy model family. Tencent positions it as a substantial step beyond Hy4 Preview, emphasizing stronger agentic capabilities and better performance on long-horizon, real-world productivity tasks. The notable new capability called out in the release listing is visual understanding.
Hy4 is aimed at workflows that require more than a single response: planning, carrying out a sequence of actions, using tools, and maintaining task context across multiple stages. Its reported 1M-token context window makes it suitable for applications that need to supply large amounts of text and image context in one request. These are positioning and API-catalog claims, not a guarantee that every long-running workflow will succeed without application-side orchestration.
Main features of Tencent Hy4
- Improved agentic execution: Tencent and the model listing describe gains in agentic capability, especially for multi-stage and long-horizon productivity work.
- Visual understanding: The official Hy4 release is described as introducing visual understanding to the Hy4 line for the first time; OpenRouter lists image input.
- Long-context processing: The listed 1,048,576-token context window is useful for large document sets, extended conversations, and codebase context.
- Coding and software workflows: Tencent positions the Hy4 series around real-world productivity, with the preview release emphasizing software engineering tasks such as planning, debugging, and validation. Do not treat preview-specific measured results as Hy4 benchmark scores.
- Tool calling: OpenRouter's API documentation lists
toolsandtool_choice, allowing applications to expose functions to the model. The application remains responsible for executing and authorizing tools. - Structured responses: OpenRouter documents JSON Schema structured output through
response_format, useful when downstream software needs predictable fields.
Tencent Hy4 vs. related Tencent models
| Model | Positioning | Context / output information found | Practical distinction |
|---|---|---|---|
| Hy4 | Official release focused on agentic and long-horizon productivity; adds visual understanding | 1M context and 65,536 max completion on CometAPI | Choose when you need the newer release's agentic focus and image understanding. |
| Hy4 Preview | Earlier preview release, focused on productivity, coding, office work, and research | Tencent's release materials describe a context window above 1M; TokenHub lists 1M context, 960K max input, and 64K max output | Use only when you specifically need the preview model/version or its documented deployment path. |
| Hy3 | Earlier Tencent model with reasoning, coding, and agent-task positioning | Tencent TokenHub's model list reports 256K context and 128K max output | A related alternative, but its smaller context window and different positioning make it a distinct choice rather than a drop-in quality equivalent. |
Recommended use cases
- Repository-scale coding assistance: Provide code, issue history, logs, and architecture context; use tool calling for controlled test and inspection workflows.
- Document-heavy analysis: Analyze long reports, policies, and cross-document material within a large context window.
- Visual document and interface analysis: Use image input for screenshots, diagrams, or visual material, subject to the capabilities of the chosen API provider.
- Multi-step office automation: Combine planning and structured outputs with application-controlled tools to create or update business documents and data.
- Agentic research workflows: Use a tool-enabled application to gather evidence, analyze it, and return a structured result; validate tool outputs and sources independently.