QwQ-32B

131K

0.63

33B

Chat

Active

QwQ-32B

Explore QwQ-32B: Alibaba's compact reasoning model delivering state-of-the-art performance in math, coding, and logical problem-solving tasks!

Try it now

Creates a chat completion

const { OpenAI } = require('openai');

const api = new OpenAI({
  baseURL: 'https://api.aimlapi.com/v1',
  apiKey: '<YOUR_API_KEY>',
});

const main = async () => {
  const result = await api.chat.completions.create({
    model: 'Qwen/QwQ-32B',
    messages: [
      {
        role: 'system',
        content: 'You are an AI assistant who knows everything.',
      },
      {
        role: 'user',
        content: 'Tell me, why is the sky blue?'
      }
    ],
  });

  const message = result.choices[0].message.content;
  console.log(`Assistant: ${message}`);
};

main();

‍

AI Playground

Test all API models in the sandbox environment before you integrate. We provide more than 200 models to integrate into your app.

Testimonials

Our Clients' Voices

Sheldon Lewis

Chief Compliance Officer (CCO)

I love the seamless integration with OpenAI. Transitioning my projects was smooth and hassle-free. Plus, the cost savings are incredible!

Will jack ds

IT Systems Manager

This groundbreaking API empowers developers with access to over 100 AI models through a single interface, fostering continuous innovation around the clock. It boasts GPT-4 level performance at a fraction of the cost, making advanced AI capabilities more accessible than ever. Seamless compatibility with OpenAI ensures smooth transitions and integration, setting a new standard for efficiency and scalability in AI development.

Oksana Kirilenko

Senior Software Engineer

AI/ML API is a promising solution for developers seeking a cost-effective and user-friendly way to integrate advanced AI features. Its extensive model library, affordability, and ease of use make it a compelling option. However, for projects requiring a wider range of user reviews or a more established platform, further research into competitors might be beneficial

QwQ-32B

Reasoning with reinforcement learning.

QwQ-32B Description

Overview:

QwQ-32B is a compact reasoning model designed to tackle complex problem-solving tasks with state-of-the-art efficiency. Despite its relatively small size of 32 billion parameters, it achieves performance comparable to much larger models like DeepSeek-R1 (671 billion parameters). Leveraging reinforcement learning (RL) and agentic capabilities, QwQ-32B excels in mathematical reasoning, coding, and structured workflows.

Key Features:

Compact yet powerful: Achieves near-parity with larger models while requiring significantly less computational power.
Reinforcement learning-driven reasoning: Integrates multi-stage RL for improved problem-solving and adaptability.
Agentic capabilities: Dynamically adjusts reasoning processes based on environmental feedback.
Wide context window: Processes up to 131,072 tokens for handling long-form inputs effectively.

Intended Use:

QwQ-32B is tailored for applications requiring structured reasoning and problem-solving, including:

Mathematical and scientific research.
Complex coding tasks and debugging.
Logical workflows in finance and engineering.
AI-powered agents requiring dynamic adaptability.

Language Support:

While specific language support details are not provided, QwQ-32B likely supports multiple languages due to its broad training data and reasoning focus.

Technical Details

Architecture:

QwQ-32B employs a transformer-based architecture with advanced components such as:

RoPE (Rotary Position Embedding): For improved token positional encoding.
SwiGLU activation functions for enhanced efficiency.
RMSNorm normalization to stabilize training.
Attention QKV biasing for better attention mechanism performance.

The model features:

Parameters: 32.5 billion total (31 billion non-embedding).
Layers: 64 transformer layers.

Training Data:

The model was trained using a combination of pretraining, supervised fine-tuning, and reinforcement learning (RL). Training data likely includes diverse datasets covering math, coding, logic, and general knowledge domains.

Diversity and Bias:

Reinforcement learning techniques were used to improve alignment with human preferences and reduce biases in responses. However, as with all AI models, residual biases may persist due to limitations in training data diversity.

Performance Metrics:

QwQ-32B has demonstrated impressive performance metrics:

Usage

Code Samples:

The model is available on the AI/ML API platform as "QwQ-32B" .

Creates a chat completion

const { OpenAI } = require('openai');

const api = new OpenAI({
  baseURL: 'https://api.aimlapi.com/v1',
  apiKey: '<YOUR_API_KEY>',
});

const main = async () => {
  const result = await api.chat.completions.create({
    model: 'Qwen/QwQ-32B',
    messages: [
      {
        role: 'system',
        content: 'You are an AI assistant who knows everything.',
      },
      {
        role: 'user',
        content: 'Tell me, why is the sky blue?'
      }
    ],
  });

  const message = result.choices[0].message.content;
  console.log(`Assistant: ${message}`);
};

main();

API Documentation:

Detailed API Documentation is available here.

Ethical Guidelines

The Qwen Team has emphasized safety by employing rule-based verifiers during training to ensure correctness in outputs for math and coding tasks. However, users should remain cautious about potential biases or inaccuracies in less-tested domains.

Licensing

QwQ-32B is open-source under the Apache 2.0 license, allowing free use for commercial and research purposes. It is deployable on consumer-grade hardware due to its compact size.

‍

Get QwQ-32B API here.

Try it now