# Qwen2.5 32B Instruct ![Organization](https://img.shields.io/badge/Org-Alibaba-ff6a00) ![License](https://img.shields.io/badge/License-Apache%202.0-22c55e) ![Released](https://img.shields.io/badge/Released-Sep%202024-555) ![Params](https://img.shields.io/badge/Params-32B-blue) > **Quick answer:** Qwen2.5 32B Instruct is Alibaba's September 2024 general-purpose 32B model scoring 88.4% HumanEval, 95.9% GSM8K, 79.9% ACEBench, and 69% MMLU-Pro. Apache 2.0 with 128K context. ## At a Glance **Where Qwen2.5 32B Instruct leads** - 95.9% GSM8K — excellent math - 88.4% HumanEval — strong coding - 79.9% ACEBench — good agent/tool use - Apache 2.0 — fully open - 32B: middle-tier efficient deployment **Where it lags** - 69% MMLU-Pro — moderate academic knowledge - 24.6% BigCodeBench — limited complex code generation - September 2024: superseded by Qwen2.5 Coder 32B and Qwen3 **Best for:** General instruction-following at 32B scale; balanced math+coding+agent deployments; Apache 2.0 mid-tier deployment. ## What Qwen2.5 32B Instruct Is Qwen2.5 32B Instruct is the general-purpose instruction-tuned 32B model in the Qwen2.5 series, released September 2024. It occupies the middle tier between Qwen2.5 7B and Qwen2.5 72B. With 95.9% GSM8K, 88.4% HumanEval, and 79.9% ACEBench, this model provides a balanced profile for math, coding, and tool-use — without the code-specialisation trade-offs of Qwen2.5 Coder 32B. The 128K context window is standard for this generation. ## Specifications | Field | Value | |---|---| | Organization | Alibaba | | License | Apache 2.0 | | HuggingFace | Qwen/Qwen2.5-32B-Instruct | | Release date | September 19, 2024 | | Parameters | 32B | | Modality | Text only | | Context window | 128K tokens | ## Pricing Open weights under Apache 2.0 — self-host. Available via Alibaba Cloud and major providers. ## Public Benchmark Scores | Benchmark | Score | Source | Date | |---|---|---|---| | [GSM8K](/benchmarks/gsm8k) | 95.9% | Benchgen evaluation | 2024-09 | | [HumanEval](/benchmarks/humaneval) | 88.4% | Benchgen evaluation | 2024-09 | | [ACEBench](/benchmarks/acebench) | 79.9% | Benchgen evaluation | 2024-09 | | [MMLU-Pro](/benchmarks/mmlu-pro) | 69% | Benchgen evaluation | 2024-09 | | [BigCodeBench](/benchmarks/bigcodebench) | 24.6% | Benchgen evaluation | 2024-09 | ## Qwen2.5 32B Instruct vs Alternatives | Model | GSM8K | HumanEval | ACEBench | License | |---|---|---|---|---| | **Qwen2.5 32B Instruct** | 95.9% | 88.4% | 79.9% | Apache 2.0 | | [Qwen2.5 Coder 32B Instruct](/models/alibaba/qwen2-5-coder-32b-instruct) | 91.1% | 92.7% | 85.3% | Apache 2.0 | | [Qwen2.5 72B Instruct](/models/alibaba/qwen2-5-72b-instruct) | 95.8% | 86.6% | — | Apache 2.0 | | [Qwen2.5 7B Instruct](/models/alibaba/qwen2-5-7b-instruct) | 91.6% | 84.8% | 57.8% | Apache 2.0 | Qwen2.5 32B vs Coder 32B: higher GSM8K (95.9% vs 91.1%), lower HumanEval (88.4% vs 92.7%), lower ACEBench (79.9% vs 85.3%). For coding agents: Coder 32B. For general balance: 32B Instruct. ## Related Models & Benchmarks - Compare: [Qwen2.5 Coder 32B Instruct](/models/alibaba/qwen2-5-coder-32b-instruct) · [Qwen2.5 72B Instruct](/models/alibaba/qwen2-5-72b-instruct) · [Qwen2.5 7B Instruct](/models/alibaba/qwen2-5-7b-instruct) - Benchmarks: [GSM8K](/benchmarks/gsm8k) · [HumanEval](/benchmarks/humaneval) · [ACEBench](/benchmarks/acebench) ## Frequently Asked Questions

What is Qwen2.5 32B Instruct?

Alibaba's September 2024 general-purpose 32B model scoring 95.9% GSM8K, 88.4% HumanEval, 79.9% ACEBench. Apache 2.0 with 128K context.

Should I use Qwen2.5 32B or Qwen2.5 Coder 32B?

Qwen2.5 Coder 32B for coding agents and SWE tasks (higher HumanEval: 92.7% vs 88.4%, higher ACEBench: 85.3% vs 79.9%). Qwen2.5 32B Instruct for general tasks with balanced math and moderate coding.

--- *Specs from Alibaba's Qwen2.5 32B Instruct release (September 2024) and Benchgen evaluations. Last updated 2026-07-24.*

Qwen2.5 32B Instruct — Evaluations

Evaluations