Tencent: Hy4 preview

Tencent: Hy4 preview is a text model from Tencent, released by the provider on 2026-08-28. It is a mixture-of-experts model with 49B active parameters out of 770B total. The preview label indicates that this is an early version of the Hy4 line, so it is worth treating as a model to evaluate rather than one to assume is final. It sits in the text category, meaning its purpose is language work rather than image, audio or other modalities.

On the input side, Hy4 preview accepts text only, and it produces text only. The context window is 1,048,576 tokens, which is large enough to hold entire codebases, long document sets or extended agent transcripts in a single request. The maximum output is 64,000 tokens, so responses can be long as well. Tencent describes the model as designed for coding agents, complex tool-use workflows, and productivity tasks, which lines up with the tool calling and structured output parameters it supports.

Those capabilities suggest a few practical uses. Coding agents can read a repository or a large slice of one and then plan edits, write patches or answer questions about the code. Tool-use workflows fit because the model accepts tool definitions and tool choice settings, so it can decide when to call a function and how to use the result. Long context also helps with document analysis, summarisation of large corpora, and multi-step productivity tasks where the model needs to keep earlier instructions and intermediate results in view. Structured outputs make it usable in pipelines that need JSON or another fixed shape rather than free text.

You call the model through api14's OpenAI-compatible API using the model id "tencent/hy4-preview". Because the interface follows the OpenAI format, existing clients and SDKs generally work with only the base URL and model id changed. The request parameters it supports are frequency_penalty, include_reasoning, logit_bias, max_completion_tokens, max_tokens, min_p, presence_penalty, reasoning, reasoning_effort, repetition_penalty, response_format, seed, stop, structured_outputs, temperature, tool_choice, tools, top_k and top_p. Reasoning controls such as reasoning_effort and include_reasoning are available, as are sampling controls, penalties, stop sequences and seeds for reproducibility.

One thing to keep in mind is that this is a preview release, so behaviour and availability may change. Input is text only, so tasks that need image or audio understanding are out of scope here. If your workload fits within a smaller context window or does not need tool use or structured outputs, a simpler text model may be a better fit. For long-context coding agents and tool-heavy workflows, though, Hy4 preview is built for exactly that shape of work.