Leaderboard / Alibaba (Qwen)

Qwen3.5 4B

Name: Qwen3.5 4B
Author: Alibaba (Qwen)

Alibaba (Qwen) · Qwen3.5 · released 2026-03-02 · data checked 2026-11-01

Qwen3.5 4B is a 4B-parameter dense model from Alibaba (Qwen) (China), released 2026-03-02 under the Apache-2.0 license. It scores 57.8 on the Small LM Index, #6 of 21 tiny models, with 1 independently measured benchmark. It needs about 3.3 GB of memory at 4-bit with an 8K-token context.

Small LM Index

57.8

#6 of 21 scored tiny models

Likely 51.8–63.3 · from 3 of 6 core benchmarks

Parameters

Dense

Price per 1M tokens

—

No hosted API on OpenRouter; self-host

Context window

262K

tokens

Benchmarks

Each bar shows this model's score. The tick marks the best tiny model on that benchmark. Sources: 5 reported by Alibaba (Qwen), 1 independently measured. How we label sources.

MMLU-Pro Lab-reported Knowledge79.1

GPQA Diamond Lab-reported Knowledge76.2

Class best: Nanbeige4.2 3B 87.4

LiveCodeBench Lab-reported Coding · LiveCodeBench v655.8

Class best: LFM3 1.7B 73.5

τ²-Bench Lab-reported Agentic79.9

Class best: MiniCPM5 2B 97.1

IFBench Lab-reported Instruction following59.2

Class best: Spark-X2.6 4B 75.8

MMMU-Pro Independent Vision · Measured by Artificial Analysis65.4

Not reported: estimated

What the index model expects from this model's other scores. Not a measurement.

Humanity's Last Exam≈ 9

AIME≈ 81

SWE-bench Verified≈ 42

Specifications

Lab: Alibaba (Qwen) (CN · China)
Architecture: Dense transformer
Total parameters: 4B
Active parameters: All (dense)
Size class: Tiny
Context window: 262K tokens
Input modalities: text, image, video
Reasoning: Hybrid (thinking on/off)
License: Apache-2.0
Openness: Open weights
Memory to run: ~3.3 GB at 4-bit, 8K context (more sizes)
Hugging Face: 6.1M downloads (30 days) · 1.2K likes

Weights on Hugging Face ↗

Compare with

Qwen3.5 4B vs Nanbeige4.2 3B (4.2B, index 70.4)
Qwen3.5 4B vs Spark-X2.6 4B (4.1B, index 67.5)
Qwen3.5 4B vs Spark-X2.5 4B (4.1B, index 65.5)
Qwen3.5 4B vs Nemotron Labs Dory 1.0 4B (4B, index 60.0)
Qwen3.5 4B vs Qwen3 4B Thinking 2507 (4B, index 55.7)
Qwen3.5 4B vs Gemma 3 4B (4.3B, index 17.5)

Ranked in

Sources

Something wrong here? Report it.

Run it locally

Uses unsloth/Qwen3.5-4B-GGUF (Q4_K_M). Swap the tag for another size: BF16, IQ4_NL, IQ4_XS, Q3_K_M, Q3_K_S, Q4_0, Q4_1, Q4_K_M….

Ollama

ollama run hf.co/unsloth/Qwen3.5-4B-GGUF:Q4_K_M

llama.cpp

llama-server -hf unsloth/Qwen3.5-4B-GGUF:Q4_K_M

Then open http://localhost:8080 for a chat window.

MLX (Mac)

pip install -U mlx-lm
mlx_lm.chat --model mlx-community/Qwen3.5-4B-MLX-4bit

Text only; for images use mlx-vlm.

vLLM (GPU server)

vllm serve Qwen/Qwen3.5-4B

Full-precision weights, for a server GPU.

Open in LM Studio

Memory needed: weights, context cache and runtime buffers.

Quantization	4K context	32K context	256K context
16-bit F16	8.7 GB	9.7 GB	17 GB
8-bit Q8_0	5.0 GB	5.9 GB	13 GB
4-bit Q4_K_M	3.1 GB	4.1 GB	12 GB

Check it against your GPU, Mac or phone →