---
title: Qwen/Qwen3.5-4B
description: Qwen/Qwen3.5-4B — Qwen3 MoE model with 4.0B parameters.
canonical_url: https://superlinked.com/models/qwen-qwen3-5-4b
last_updated: 2026-06-29
---

# Qwen/Qwen3.5-4B

Source: [Qwen/Qwen3.5-4B on HuggingFace](https://huggingface.co/Qwen/Qwen3.5-4B)

## Overview

| Field | Value |
|-------|-------|
| Architecture | Qwen3 MoE |
| Parameters | 4.0B |
| Tasks | Generate |
| Outputs | Text |
| Context window | 8,192 tokens |
| Max output | 4,096 tokens |
| Capabilities | tool calling, constrained output (JSON Schema, Regex), streaming |
| License | apache-2.0 |
| Inputs | text, image |

## Benchmarks

### CaseHOLD

Domain: legal · Task: generation · Language: en

**Quality:** accuracy: 0.5867

**Performance (RTX-PRO-6000 b1 c4):** Corpus 234 tok/s · Corpus p50 788.1ms

### GPQA Diamond

Domain: scientific · Task: generation · Language: en

**Quality:** accuracy: 0.4495

**Performance (RTX-PRO-6000 b1 c4):** Corpus 343 tok/s · Corpus p50 863.8ms

### MedQA

Domain: medical · Task: generation · Language: en

**Quality:** accuracy: 0.6700

**Performance (RTX-PRO-6000 b1 c4):** Corpus 364 tok/s · Corpus p50 735.2ms

### MMLU-Pro

Domain: general · Task: generation · Language: en

**Quality:** accuracy: 0.5767

**Performance (RTX-PRO-6000 b1 c4):** Corpus 390 tok/s · Corpus p50 587.2ms