Model Details
Qwen2.5 32B Instruct
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- September 2024: superseded by Qwen2.5 Coder 32B and Qwen3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 | 95.9% | Benchgen evaluation | 2024-09 |
| HumanEval | 88.4% | Benchgen evaluation | 2024-09 |
| ACEBench | 79.9% | Benchgen evaluation | 2024-09 |
| MMLU-Pro | 69% | Benchgen evaluation | 2024-09 |
| 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 | 91.1% | 92.7% | 85.3% | Apache 2.0 |
| Qwen2.5 72B Instruct | 95.8% | 86.6% | — | Apache 2.0 |
| 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 · Qwen2.5 72B Instruct · Qwen2.5 7B Instruct
- Benchmarks: GSM8K · HumanEval · 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.
Benchmark Leaderboards
This model isn’t on any benchmark leaderboard yet.