Model Details
Qwen3 VL 30B A3B Instruct
Quick answer: Qwen3 VL 30B A3B Instruct is Alibaba's February 2026 compact MoE vision-language instruct model scoring 77.8% MMLU-Pro, 66.3% BFCL-v3, and 58.5% Arena Hard v2. Apache 2.0 — 3B active params.
At a Glance
Where Qwen3 VL 30B A3B Instruct leads
- 77.8% MMLU-Pro — strong knowledge at 3B active
- 66.3% BFCL-v3 — good function calling
- Apache 2.0 — fully open
- 3B active: cost-efficient VL instruct
Where it lags
- Lower than 30B A3B Thinking on MMLU-Pro (77.8% vs 80.5%)
- Non-thinking: less reasoning depth
Best for: Cost-efficient VL instruct deployments; faster inference than thinking variant.
What Qwen3 VL 30B A3B Instruct Is
The standard instruct (non-thinking) 30B A3B MoE variant. Faster inference than the thinking variant, with slightly lower quality.
Specifications
| Field | Value |
|---|---|
| Organization | Alibaba |
| License | Apache 2.0 |
| HuggingFace | Qwen/Qwen3-VL-30B-A3B-Instruct |
| Release date | February 2026 |
| Parameters | 30B total / 3B active (MoE) |
| Modality | Text and vision |
Pricing
Open weights under Apache 2.0 — self-host at no cost.
Public Benchmark Scores
| Benchmark | Score | Source | Date |
|---|---|---|---|
| MMLU-Pro | 77.8% | Benchgen evaluation | 2026-02 |
| BFCL-v3 | 66.3% | Benchgen evaluation | 2026-02 |
| Arena Hard v2 | 58.5% | Benchgen evaluation | 2026-02 |
Qwen3 VL 30B A3B Instruct vs Alternatives
| Model | MMLU-Pro | BFCL-v3 | Arena Hard | Type |
|---|---|---|---|---|
| Qwen3 VL 30B A3B Instruct | 77.8% | 66.3% | 58.5% | MoE Instruct |
| Qwen3 VL 30B A3B Thinking | 80.5% | 68.6% | 56.7% | MoE Thinking |
| Qwen3 VL 32B Instruct | 78.6% | 70.2% | 64.7% | Dense Instruct |
For maximum Arena Hard: 32B Instruct (64.7%). For MMLU-Pro at 3B active: 30B A3B Thinking (80.5%).
Related Models & Benchmarks
- Compare: Qwen3 VL 30B A3B Thinking · Qwen3 VL 32B Instruct
- Benchmarks: MMLU-Pro · BFCL-v3 · Arena Hard v2
Frequently Asked Questions
What is Qwen3 VL 30B A3B Instruct?
Alibaba's February 2026 compact MoE VL instruct model (30B/3B active) scoring 77.8% MMLU-Pro, 58.5% Arena Hard. Apache 2.0.Specs from Alibaba's Qwen3 VL 30B A3B Instruct release (February 2026) and Benchgen evaluations. Last updated 2026-07-24.
Benchmark Leaderboards
This model isn’t on any benchmark leaderboard yet.