Qwen: Qwen3 235B A22B

Inference Model ID: qwen/qwen3-235b-a22b

Overview

Description

Qwen3-235B-A22B is a 235 billion parameter mixture-of-experts (MoE) model developed by Qwen (Alibaba Cloud), activating 22 billion parameters per forward pass. This architecture allows the model to deliver exceptional performance while maintaining computational efficiency.

The model supports seamless switching between two operational modes:

• "Thinking" Mode: For complex reasoning, mathematics, and code tasks - uses explicit reasoning tokens (<think> and </think>)
• "Non-thinking" Mode: For general conversational efficiency with faster response times

Key characteristics:

• Strong Reasoning Ability: Excels at mathematical problem-solving and complex logical tasks
• Multilingual Support: Supports 100+ languages and dialects
• Advanced Instruction-Following: High accuracy in following complex instructions
• Agent Tool-Calling: Native support for tool/function calling capabilities
• Extended Context: Native 32K context window, extendable to 131K tokens using YaRN-based scaling

Pricing

Cost Comparison

• This model offers competitive pricing for a 235B parameter MoE model
• The MoE architecture (22B active parameters) enables lower per-token costs compared to dense models of similar capability

Supported Parameters

Default Configuration

• Default Stop Tokens: <|im_start|>, <|im_end|>
• Instruction Type: Qwen3

Features

• Reasoning Tokens: Uses <think> and </think> tokens for explicit reasoning
• Tool Calling: Native support for function/tool calling
• Prompt Caching: Supported for repeated requests
• Multipart Input: Supports multipart message format

Performance

Architecture Details

Recent Usage Statistics

Capabilities

• Reasoning: Strong performance on mathematical and logical reasoning benchmarks
• Code Generation: Proficient in multiple programming languages
• Multilingual: Supports 100+ languages and dialects
• Long Context: Effective utilization of extended context windows

API Usage Example

curl -X POST https://api.langmart.ai/v1/chat/completions \
  -H "Authorization: Bearer $LANGMART_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "qwen/qwen3-235b-a22b",
    "messages": [
      {"role": "user", "content": "Explain the concept of mixture-of-experts architecture in neural networks."}
    ],
    "max_tokens": 2048,
    "temperature": 0.7
  }'

Reasoning Mode Example

curl -X POST https://api.langmart.ai/v1/chat/completions \
  -H "Authorization: Bearer $LANGMART_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "qwen/qwen3-235b-a22b",
    "messages": [
      {"role": "user", "content": "Solve this step by step: If a train travels 120 km in 2 hours, then stops for 30 minutes, then travels 90 km in 1.5 hours, what is the average speed for the entire journey?"}
    ],
    "reasoning": true,
    "include_reasoning": true,
    "max_tokens": 4096
  }'

Tool Calling Example

curl -X POST https://api.langmart.ai/v1/chat/completions \
  -H "Authorization: Bearer $LANGMART_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "qwen/qwen3-235b-a22b",
    "messages": [
      {"role": "user", "content": "What is the weather in Tokyo?"}
    ],
    "tools": [
      {
        "type": "function",
        "function": {
          "name": "get_weather",
          "description": "Get current weather for a location",
          "parameters": {
            "type": "object",
            "properties": {
              "location": {"type": "string", "description": "City name"}
            },
            "required": ["location"]
          }
        }
      }
    ],
    "tool_choice": "auto"
  }'

Related Models

Qwen3 Model Family

Similar MoE Models

Providers

Primary Provider: DeepInfra

Notes

• This model is part of Qwen's third-generation model family released in April 2025
• The MoE architecture (235B total, 22B active) provides an excellent balance of capability and efficiency
• Supports both "thinking" mode for complex tasks and "non-thinking" mode for quick responses
• YaRN-based context extension allows handling documents up to 131K tokens
• Currently available through DeepInfra with FP8 quantization
• Native support for tool calling makes it suitable for agent-based applications

Source: LangMart Model Registry Last Updated: December 23, 2025