Tencent: Hy-MT2-30B-A3B

Tencent: Hy-MT2-30B-A3B is a text model from Tencent, positioned as the flagship translation model in the Hy-MT2 family. It is a text-in, text-out model, so it takes prompts as text and returns text, with no image or audio handling. If you are building translation or multilingual text processing into an application, this is the model in the family that Tencent treats as its top option.

The model supports 33 language pairs along with five Chinese dialect and minority-language pairs. Beyond plain sentence translation, the provider describes workflows for structured, delimiter-based, contextual, and glossary-based translation, which means you can shape the output format and supply terminology that the model should follow. The context window is 8192 tokens, and the maximum output is 4096 tokens, so a single request can hold a reasonably long source passage plus instructions while still leaving room for the translated result. Inputs and outputs are text only.

Those capabilities point to a fairly specific set of use cases. Localization pipelines are the obvious fit: translating product strings, documentation, or support content across many language pairs, with glossary support helping keep product names and domain terms consistent. Delimiter-based and structured workflows suit batch jobs where you send several segments in one request and need the responses separated in a predictable way. Contextual translation helps when a sentence cannot be translated well in isolation, such as dialogue, UI copy with placeholders, or text that refers back to earlier sentences. The Chinese dialect and minority-language pairs are useful for regional products and content that mainstream translation models often handle poorly. Because the output limit is 4096 tokens, long documents are better split into chunks than sent whole.

You call the model through api14's OpenAI-compatible API using the model id "tencent/hy-mt2-30b-a3b". Because the interface follows the OpenAI chat completions shape, you can point an existing OpenAI client at api14's base URL and change the model id, with no other structural changes. The supported request parameters are max_completion_tokens, max_tokens, response_format, stop, structured_outputs, and temperature. Use temperature to control how literal or varied the translation is, stop to end generation at a chosen sequence, and response_format or structured_outputs when you need the reply in a defined shape rather than free text. Note that max_tokens and max_completion_tokens are both accepted, so pick one convention and stay consistent.

The main constraint to plan around is size. With an 8192 token context window and a 4096 token output ceiling, this model is built for passages and segments rather than whole books in one call. If your task is general reasoning, code generation, or long-document analysis rather than translation, a general purpose text model will usually fit better. Within translation work, though, the language coverage, dialect pairs, and glossary and delimiter workflows are the reasons to choose this model over a generic one.