Qwen2-Math
Math-specialist LLMs in 1.5B, 7B, and 72B that beat frontier models on math.
About
Qwen2-Math, released by Alibaba in August 2024, is a series of mathematics-specialist language models built on the Qwen2 base. The 1.5B, 7B, and 72B models were pretrained on a curated math corpus of articles, books, exams, and synthetic data, then instruction-tuned with a math-specific reward model and Group Relative Policy Optimization (GRPO).
The 72B instruct model outperformed GPT-4o, Claude 3.5 Sonnet, Gemini 1.5 Pro, and Llama 3.1 405B on math benchmarks including MATH, GSM8K, OlympiadBench, and AIME 2024, a striking result for an open-weight specialist. License nuance: smaller sizes are openly licensed, but the 72B carries commercial-use conditions for large-scale deployers, so check the model card before commercial use. Superseded in part by Qwen2.5-Math, but the series that proved open specialists could beat closed generalists at math.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 8/10
Scored 8/10 (verified tier, metrics-v1). Evidence: Verified via vendor site/docs/GitHub and reputable coverage, wave 78; Trending score 78/100 from researcher assessment.