Alibaba: qwen-math-plus-latest API

Qwen-math-plus is a specialized large language model optimized for advanced mathematical reasoning and scientific problem-solving. Unlike general-purpose models, it is pre-trained on an extensive corpus of mathematical web texts, textbooks, and synthetic data (TIR/CoT). The "Plus" version integrates a native Python Interpreter (Tool-Integrated Reasoning, TIR), allowing it to verify calculations in real-time. In 2026, the model features an optional "Thinking Mode" which utilizes large-scale reinforcement learning to perform systematic self-correction during the problem-solving process. It consistently outperforms GPT-4o and deepseek-v3 in competitive math benchmarks such as MATH, GSM8K, and AIME 2024/2025.

• Context window: 4,096 tokens
• Max output: 4,096 tokens
• Input: text
• Output: text
• Released: 2025-09-18
• Knowledge cutoff: 2025-10-31