Models / Tencent
Chat
Reasoning
Vision

Tencent Hy4

Vision and computer use for professional workflows, with 1M context

About model

Tencent Hy4 has 1M-token context window. On Together AI it costs $0.83 per 1M input tokens and $2.50 per 1M output tokens.

Hy4 is Tencent's flagship model and the successor to Hy4 Preview, adding visual understanding to the Hy family for the first time along with computer use. It interprets charts, documents, and detailed images, reasons across multiple images, and solves math problems that depend on visual information. As an agent, it works through both command-line tools and graphical interfaces, operating professional software such as Blender and CAD applications and carrying out multi-step tasks that span several programs. Tencent developed the model with its product teams, training and evaluating it against real workflows in office work, financial analysis, and game development, and reports large reductions in speculative errors compared with the preview. The model supports a 1M-token context window with adjustable reasoning effort, tool calling, and structured output. Available on Together AI.

‍

Context Window

1M

Long agent sessions, large projects, and document sets in one request

Vision Input

Native

Charts, documents, and multi-image reasoning, new to the Hy family

Computer Use

CLI + GUI

Operates command-line tools and graphical software such as Blender and CAD

Model key capabilities
  • Computer Use: Works through command-line tools and graphical interfaces, operating professional software such as Blender and CAD across multi-program tasks
  • Visual Understanding: Interprets charts, documents, and detailed images, reasons across multiple images, and solves visual math problems
  • Professional Workflows: Trained against real workflows in office work, financial research, and game development across Unreal Engine, Unity, and Godot
  • Production-Ready Infrastructure: 99.9% SLA, available on serverless and dedicated infrastructure
  • API usage

    • cURL
    • Python
    • Typescript

    Endpoint:

    tencent/Hy4

    curl -X POST "https://api.together.ai/v1/chat/completions" \
      -H "Authorization: Bearer $TOGETHER_API_KEY" \
      -H "Content-Type: application/json" \
      -d '{
        "model": "tencent/Hy4",
        "messages": [
          {
            "role": "user",
            "content": "What are some fun things to do in New York?"
          }
        ]
    }'
    
    from together import Together
    
    client = Together()
    
    response = client.chat.completions.create(
      model="tencent/Hy4",
      messages=[
        {
          "role": "user",
          "content": "What are some fun things to do in New York?"
        }
      ]
    )
    print(response.choices[0].message.content)
    
    import Together from 'together-ai';
    const together = new Together();
    
    const completion = await together.chat.completions.create({
      model: 'tencent/Hy4',
      messages: [
        {
          role: 'user',
          content: 'What are some fun things to do in New York?'
         }
      ],
    });
    
    console.log(completion.choices[0].message.content);
    
  • Model card

    Architecture Overview:
    • Successor to Hy4 Preview, Tencent's open-weight Mixture-of-Experts model from August 2026; architecture details for this release arrive with Tencent's system card
    • Text and image input with text output, across a 1M-token context window
    • Adjustable reasoning effort
    • Served on Together AI from the NVFP4 checkpoint

    Training Methodology:
    • Developed with Tencent product teams, who trained and evaluated the model against real workflows
    • Targeted training for computer use, including Blender, CAD software, and tasks spanning multiple programs
    • Financial training built on thousands of specialized workspaces and tens of thousands of professional tasks
    • Cybersecurity training limited to defensive material such as vulnerability analysis and remediation, with safeguards targeting cybersecurity and biological risks

    Performance Characteristics:
    • All figures Tencent-reported against Hy4 Preview unless noted
    • Errors from unsupported assumptions: 5.6%, down from 23.1%
    • Game development: GameDev up 19.1% and GameLogic up 9.8%
    • Financial research: average score of 3.04 in Tencent's blind evaluation by 213 professionals across 579 real-world tasks
    • Tencent notes the model still trails top proprietary models on its hardest internal tests, including research-level mathematics and long-horizon coding

    ‍

  • Prompting

    Together AI API Access:
    • Access Tencent Hy4 via Together AI APIs using the endpoint tencent/Hy4
    • Authenticate using your Together AI API key in request headers
    • Supports text and image input across a 1M-token context window, with adjustable reasoning effort
    • Supports tool calling and structured output for agent workflows
    • Available on Together AI serverless and dedicated infrastructure

    ‍

  • Applications & use cases

    Computer Use & Automation:
    • Automate multi-step tasks across command-line tools and desktop software
    • Build 3D scenes and models through Blender and CAD
    • Produce documents and web pages end to end

    Finance & Office Work:
    • Run fund research, securities analysis, and quantitative strategy work
    • Track projects, flag unfinished assignments, and surface approaching deadlines
    • Read charts and documents as part of analysis workflows

    Game Development:
    • Build game mechanics and multiplayer features in Unreal Engine, Unity, and Godot
    • Plan and implement multi-step development tasks
    • Generate assets and scenes for interactive projects

    ‍

  • Model provider
    Tencent
  • Type
    Chat
    Reasoning
    Vision
  • Features
    Function Calling
    JSON Mode
  • Endpoint
  • Context length
    1M
  • Input price

    $0.83 / 1M tokens

    $0.042 (cached)/1M

  • Output price

    $2.50 / 1M tokens

  • Input modalities
    Text
    Image
  • Output modalities
    Text
  • Released
    November 18, 2026
  • Quantization level
    NVFP4
  • Category
    Chat