Skip to main content
Kaino.dev
Discover
Evals
News
Academics
Insights
Kaino.dev

Discover, evaluate, and compare AI tools, models, and agents.

Explore

  • Discover
  • Evaluations
  • News
  • Academics
  • Insights

Community

  • Twitter
  • YouTube
  • Instagram
Privacy PolicyTerms of Service

© 2026 Kaino.dev. All rights reserved.

Version 1.1.0
Tencent: Hy3 preview · Discover · Kaino
Discover/MODELS/Tencent: Hy3 preview
Tencent: Hy3 preview logo

MODELS

Tencent: Hy3 preview

by Tencent

modellead-sourceopenrouter-modelssource:github.comTencentHunyuanHy3 previewMixture-of-Expertsreasoningagentsopen-sourcelong-context
GitHub

Overview

Hy3 preview is Tencent’s open-sourced 295B-parameter MoE reasoning and agent model with 21B activated parameters and a 256K context window.

Details

Hy3 preview is described by Tencent as an open-sourced 295B-parameter Mixture-of-Experts model with 21B activated parameters, built for reasoning, agent capabilities, and real-world usability. OpenRouter describes it as a high-efficiency MoE model for agentic workflows and production use with configurable reasoning levels: disabled, low, and high. Tencent says the model is available through GitHub, Hugging Face, and Tencent Cloud TokenHub, and the GitHub repository provides quickstart deployment examples for vLLM and SGLang using the tencent/Hy3-preview model.

When to Use

Use for evaluating Tencent’s latest open-sourced MoE reasoning model in agentic workflows. Use when a long-context model is needed, since Tencent’s announcement cites a 256K context window. Use when you want to test configurable reasoning behavior through disabled, low, or high reasoning modes as described by OpenRouter.

Getting Started

  1. Read Tencent’s official announcement for the model overview, availability, context length, and pricing claims.
  2. Visit the Tencent-Hunyuan/Hy3-preview GitHub repository to access model links and quickstart examples.
  3. Run the repository’s vLLM or SGLang deployment examples using the tencent/Hy3-preview model.
  4. If using hosted access, review OpenRouter or Tencent Cloud TokenHub availability and pricing before production use.

Key Features

  • •295B-parameter Mixture-of-Experts architecture with 21B activated parameters, according to Tencent.
  • •256K context window, according to Tencent’s announcement.
  • •Configurable reasoning levels: disabled, low, and high, according to OpenRouter.
  • •GitHub repository includes quickstart deployment examples for vLLM and SGLang.
  • •Available via GitHub, Hugging Face, and Tencent Cloud TokenHub, according to Tencent.

Capabilities

  • •reasoning
  • •agent workflows
  • •long-context processing
  • •configurable reasoning modes
  • •vLLM deployment
  • •SGLang deployment

Last updated May 31, 2026