Model Details
Qwen2.5 VL 32B Instruct
Quick answer: Qwen2.5 VL 32B Instruct is Alibaba's March 2025 vision-language model scoring 91.5% HumanEval, 82.2% MATH, and 68.8% MMLU-Pro. Apache 2.0 — 32B multimodal.
At a Glance
Where Qwen2.5 VL 32B Instruct leads
- 91.5% HumanEval — excellent coding with vision input
- 82.2% MATH — strong math at 32B VL scale
- Apache 2.0 — fully open
- Multimodal: text + vision understanding
Where it lags
- 68.8% MMLU-Pro — moderate academic breadth
- March 2025: superseded by Qwen3 VL series
Best for: Open-source VL coding and math pipelines; Apache 2.0 vision-language 32B; image-understanding + math tasks.
What Qwen2.5 VL 32B Instruct Is
Qwen2.5 VL 32B Instruct is the 32B vision-language model from Alibaba's Qwen2.5 VL generation (March 2025). It extends Qwen2.5 32B with visual understanding, while maintaining strong HumanEval (91.5%) and MATH (82.2%) performance.
Specifications
| Field | Value |
|---|---|
| Organization | Alibaba |
| License | Apache 2.0 |
| HuggingFace | Qwen/Qwen2.5-VL-32B-Instruct |
| Release date | March 6, 2025 |
| Parameters | 32B |
| Modality | Text and vision |
Pricing
Open weights under Apache 2.0 — self-host at no cost.
Public Benchmark Scores
| Benchmark | Score | Source | Date |
|---|---|---|---|
| HumanEval | 91.5% | Benchgen evaluation | 2025-03 |
| MATH | 82.2% | Benchgen evaluation | 2025-03 |
| MMLU-Pro | 68.8% | Benchgen evaluation | 2025-03 |
Qwen2.5 VL 32B vs Alternatives
| Model | HumanEval | MATH | MMLU-Pro | Multimodal |
|---|---|---|---|---|
| Qwen2.5 VL 32B Instruct | 91.5% | 82.2% | 68.8% | Yes |
| Qwen2.5 32B Instruct | 88.4% | — | — | No |
| Qwen3 VL 32B Instruct | — | — | 78.6% | Yes |
Qwen2.5 VL 32B leads on HumanEval (91.5%) among 32B VL models. For higher MMLU-Pro: Qwen3 VL 32B Instruct (78.6%).
Related Models & Benchmarks
- Compare: Qwen3 VL 32B Instruct · Qwen2.5 32B Instruct
- Benchmarks: HumanEval · MATH · MMLU-Pro
Frequently Asked Questions
What is Qwen2.5 VL 32B Instruct?
Alibaba's March 2025 vision-language 32B model scoring 91.5% HumanEval, 82.2% MATH. Apache 2.0 — strong coding + VL.Specs from Alibaba's Qwen2.5 VL 32B Instruct release (March 2025) and Benchgen evaluations. Last updated 2026-07-24.
Benchmark Leaderboards
This model isn’t on any benchmark leaderboard yet.