Tencent: Hy3

Tencent: Hy3 is a text model created by Tencent. It is a Mixture-of-Experts model with 295 billion parameters in total, of which 21 billion are active at any time, spread across 192 experts with top-8 routing. Tencent built it for reasoning, agentic workflows, and real-world production use. The provider release date is 2026-07-06.

The model works with text input and produces text output. It has a context window of 262,144 tokens and can generate up to 128,000 tokens in a single response, so long documents, extended conversations, and large tool outputs can fit in one request. A notable capability is configurable reasoning effort, which lets you decide how much internal reasoning the model performs before answering.

Those capabilities point to a few practical uses. The large context suits analysis of lengthy documents, codebases, or transcripts in a single pass. Reasoning support fits tasks that need step-by-step problem solving, such as multi-step calculations, planning, or debugging. Tool calling and structured outputs make it a reasonable choice for agentic workflows, where the model calls functions or external services and returns results in a fixed schema. Configurable reasoning effort is useful when you want to trade depth of thinking against latency on a per-request basis.

You call the model through api14's OpenAI-compatible API using the model id "tencent/hy3". Because the interface follows the OpenAI format, existing clients and SDKs work with only the base URL, key, and model id changed. The request parameters supported 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 can be surfaced with include_reasoning and tuned with reasoning_effort, while structured_outputs and response_format cover schema-constrained replies.

On limits, the model handles only text input and text output, so it is not the right pick when you need to process images, audio, or other modalities. The 128,000-token cap on output is generous but still finite, and very long generations may be better split across requests. If you do not need reasoning or agentic tool use, a smaller and simpler text model may be a more direct fit; Hy3 makes most sense when its long context, reasoning control, and tool support are actually part of the task.