⚡️ FlashAttention-4: up to 1.3× faster than cuDNN on NVIDIA Blackwell →

Introducing Together AI's new look →

🔎 ATLAS: runtime-learning accelerators delivering up to 4x faster LLM inference →

⚡ Together GPU Clusters: self-service NVIDIA GPUs, now generally available →

📦 Batch Inference API: Process billions of tokens at 50% lower cost for most models →

🪛 Fine-Tuning Platform Upgrades: Larger Models, Longer Contexts →

Models / Qwen
Chat

Qwen3 4B

4.0B-parameter compact conversational AI model with grouped-query attention optimized for efficient chat applications and instruction following tasks.

Deploy Qwen3 4B
read docs

About model

Qwen3-4B is a 4B-parameter causal language model offering advanced reasoning, instruction-following, and multilingual support capabilities. It excels in tasks requiring complex logical reasoning, math, and coding, while also providing efficient general-purpose dialogue. Suitable for developers and researchers, Qwen3-4B supports 100+ languages and dialects, making it an ideal choice for applications requiring strong multilingual capabilities.

To run this model you first need to deploy it on a Dedicated Endpoint.

  • Model card

    Conversation Format:
    • Advanced system/user/assistant format with dynamic expert activation
    • Supports complex multi-turn dialogues with reasoning chains
    • Efficient inference through mixture-of-experts architecture
    • Strong performance on coding, mathematics, and creative tasks

    Expert Utilization:
    • Different experts activated based on input content and task requirements
    • Seamless switching between mathematical, coding, and linguistic experts
    • Contextual understanding with efficient resource allocation
    • Maintains conversation quality while optimizing computational efficiency

    Optimization Strategies:
    • Leverages specialized experts for domain-specific tasks
    • Benefits from explicit task specification in prompts
    • Responds well to structured reasoning requests
    • Optimized for both creative and analytical applications

  • Prompting

    Chat model with system/user/assistant format. Supports conversational context and instruction following capabilities.

  • Applications & use cases

    Efficient chatbots mobile assistants resource-constrained chat applications simple conversation tasks educational tools.

Related models
new
Chat
NVIDIA Nemotron 3 Super
  • Parameters
    120B
  • Deployment
    Dedicated Inference
No items found.
Code
Chat
new
Chat
MiniMax M2.5
  • Input/1M
    $0.30
  • Output/1M
    $1.20
  • Parameters
    228.7B
Function Calling
JSON Mode
Reasoning
Code
New
Chat
Kimi K2.5
  • Input/1M
    $0.50
  • Output/1M
    $2.80
  • Parameters
    1T
No items found.
Chat
Code
LLM
Vision
Model specifications
Model data
  • Model provider
    Qwen
  • Type
    Chat
  • Main use cases
    Chat
    Small & Fast
  • Fine tuning
    Supported
  • Deployment
    On-Demand Dedicated
    Monthly Reserved
  • Parameters
    4.0B
  • Context length
    32K
  • Input modalities
    Text
  • Output modalities
    Text
Quickstart docs
Deploy model