Qwen3 4B Base — Hardware Requirements & GPU Compatibility
Chat
Qwen3-4B-Base is Alibaba's 4-billion-parameter base pretrained model from the Qwen3 generation, the successor to Qwen2.5. Like other -Base checkpoints, it is a raw causal language model meant for fine-tuning or research, not direct conversation. Qwen3 was trained on 36 trillion tokens across 119 languages, tripling Qwen2.5's language coverage, using a three-stage pipeline that builds general knowledge, then reasoning skills in code and STEM, then extends context length. At 4B dense parameters, it runs easily on a single consumer GPU, even without heavy quantization. The model supports a 32,768 token context window. It is released under the Apache 2.0 license, permitting unrestricted commercial and research use, and was published in April 2025. Qwen3 introduced a mix of dense and mixture-of-experts models alongside refinements like QK layernorm for training stability.
Runs great on a 24 GB GPU
Runs from 8 GB (Q4_K_M · 32K)Best on 24 GB (Q4_K_M · ~226 tok/s · 32K)Macs from 8 GB (Q4_K_M · ~25 tok/s)
Apache 2.0
Reference GPU · RTX 4090 24 GB
Estimates, ±20% · how we estimate
Run Qwen3 4B Base
One recommended way. Every other method, OS and quant is in the step-by-step guide.
Recommended
ollama run hf.co/mradermacher/Qwen3-4B-Base-GGUF:Q4_K_MQ4_K_M fits fully on the RTX 4090 · 19.8 GB free · 16K context
Download: Qwen3-4B-Base.Q4_K_M.gguf2.50 GBmradermacher
Other ways
each opens step-by-stepRun in cloud
Fits on RTX 3060 12GB (7 GB headroom) · Q4_K_M
- Generation speed
- ~81 tok/s
- generation speed
- Cost per 1M output tokens
- $0.29
- per 1M output tokens
How Much VRAM Does Qwen3 4B Base Need?
Select a quantization level; the listed hardware adheres to it.
| Quantization | VRAM | +32K | On RTX 4090 (reference) | |
|---|---|---|---|---|
| 2.2 GB | 5 GB | 2-bit · K-quant — 2-bit quantization with K-quant improvements | Fits: 32K context · ~298 tok/s | |
| 2.3 GB | 5.1 GB | 3-bit small — 3-bit small quantization | Fits: 32K context · ~291 tok/s | |
| 2.5 GB | 5.3 GB | 3-bit medium — 3-bit medium quantization | Fits: 32K context · ~267 tok/s | |
| 2.5 GB | 5.3 GB | 4-bit · legacy — 4-bit legacy quantization | Fits: 32K context · ~262 tok/s | |
| 2.9 GB | 5.7 GB | 4-bit medium — 4-bit medium quantization — most popular sweet spot | Fits: 32K context · ~226 tok/s | |
| 3.4 GB | 6.2 GB | 5-bit medium — 5-bit medium quantization — good quality/size tradeoff | Fits: 32K context · ~196 tok/s | |
| 3.8 GB | 6.6 GB | 6-bit, very good — 6-bit quantization, very good quality | Fits: 32K context · ~172 tok/s | |
| 4.5 GB | 7.3 GB | 8-bit, near-lossless — 8-bit quantization, near-lossless | Fits: 32K context · ~145 tok/s |
Selected Q4_K_M
4-bit medium quantization4.2GBat 16K context
- Weights
- 2.4 GB
- KV cache
- 1.5 GB
- Overhead
- 0.3 GB
Fits RTX 4090 (reference) · 19.8 GB free · ~226 tok/s
≥50 tok/s needs ≥223 GB/s · RTX 5090 ~402, RTX 3090 Ti ~226
Which Hardware Can Run Qwen3 4B Base?
Q4_K_M · 2.9 GBQwen3 4B Base (Q4_K_M) requires 2.9 GB of VRAM to load the model weights. For comfortable inference with headroom for KV cache and system overhead, 4+ GB is recommended. Using the full 33K context window can add up to 2.8 GB, bringing total usage to 5.7 GB. 52 GPUs can run it, including NVIDIA GeForce RTX 5090, NVIDIA GeForce RTX 3090 Ti. 163 devices with unified memory can run Qwen3 4B Base, including Mac Studio (M3 Ultra, 256GB), Mac Studio (M3 Ultra, 512GB).
≤ 11 GB
≤ 11 GB tier →fits · 3.8 GB free · up to 32K ctx
16 GPUs · 1 Mac class · 2 device families
RTX 3080~170Simulate output on RTX 3080RTX 2080 Ti~138Simulate output on RTX 2080 TiRTX 3070 Ti~136Simulate output on RTX 3070 TiM3*8 GB~25Simulate output on iMac 24" M3 (8 GB)Orin Nano~23Simulate output on NVIDIA Jetson Orin Nano 8GB (Super)iPhone~17Simulate output on iPhone 1720 GB
fits · 15.8 GB free · up to 32K ctx
1 GPU
RX 7900 XT~166Simulate output on RX 7900 XT32–40 GB
fits · 27.8 GB free · up to 32K ctx
1 GPU · 3 Mac classes · 4 device families
RTX 5090~402Simulate output on RTX 5090M5 Max36 GB~111Simulate output on Mac Studio M5 Max (36 GB)M532 GB~37Simulate output on MacBook Air 13" M5 (32 GB)Orin~46Simulate output on NVIDIA Jetson AGX Orin 32GBSnapdragon X Elite (X1E-80-100)~30Simulate output on Snapdragon X Elite Copilot+ PC
By VRAM tier · typical GPUs and Macs at Q4_K_M
Alternatives to Qwen3 4B Base
Qwen 3 family · 77 models →Similar models
6| Model | Score | VRAM | tok/s | Why it is similar |
|---|---|---|---|---|
| Gemma 3 4B ITGoogle | 64.0 | 5 GB | ~231 | same size |
| Granite Vision 4.1 4BIBM | — | 4 GB | ~228 | same size |
| MiniCPM V 4OpenBMB | — | 3.1 GB | ~236 | same size |
| Phi 4 Mini ReasoningMicrosoft | — | 4.8 GB | ~228 | same size |
| Phi 3.5 Vision InstructMicrosoft | — | 9.2 GB | ~183 | same size |
| Nanbeige4.2 3BNanbeige | — | 3.5 GB | ~227 | same size |
Fine-tunes & variants(1)
Specifications
- Publisher
- Alibaba
- Family
- Qwen 3
- Parameters
- 4.0B
- Architecture
- Qwen3ForCausalLM
- Context Length
- 32,768 tokens
- Vocabulary Size
- 151,936
- Release Date
- 2025-04-28
- License
- Apache 2.0
Frequently Asked Questions
How much VRAM does Qwen3 4B Base need?
Qwen3 4B Base requires 2.9 GB of VRAM at Q4_K_M, or 8.5 GB at BF16. Full 33K context adds up to 2.8 GB (5.7 GB total).
VRAM = Weights + KV Cache + Overhead
Weights = 4.0B × 4.8 bits ÷ 8 = 2.4 GB
KV Cache + Overhead ≈ 0.5 GB (at 2K context + ~0.3 GB framework)
Fit ratings and hardware model lists check this model with room for a 16K-token context, which needs a little more memory.
KV Cache + Overhead ≈ 3.3 GB (at full 33K context)
VRAM usage by quantization
What's the best quantization for Qwen3 4B Base?
For Qwen3 4B Base, Q4_K_M (2.9 GB) offers the best balance of quality and VRAM usage. Q5_0 (3 GB) provides better quality if you have the VRAM. The smallest option is IQ2_XXS at 1.6 GB.
VRAM requirement by quantization
★ Recommended — best balance of quality and VRAM usage.
Can I run Qwen3 4B Base on a Mac?
Yes — MacBook Air 13" M3 (8 GB) and 145 other Macs can run Qwen3 4B Base. Apple Silicon uses unified memory, so the model shares RAM with the system. At Q4_K_M you need at least 2.9 GB of usable unified memory (RAM minus macOS overhead).
Can I run Qwen3 4B Base locally?
Yes — Qwen3 4B Base can run locally on consumer hardware. Popular tools include Ollama, LM Studio, and llama.cpp.
How fast is Qwen3 4B Base?
At Q4_K_M, Qwen3 4B Base can reach ~1655 tok/s on AMD Instinct MI350X. On NVIDIA GeForce RTX 4090: ~226 tok/s. Speed depends mainly on GPU memory bandwidth. Real-world results typically within ±20%.
tok/s = (bandwidth GB/s ÷ model GB) × efficiency
Example: NVIDIA B200 → 8000 ÷ 2.9 × 0.65 = ~1793 tok/s
Estimated speed at Q4_K_M (2.9 GB)
Real-world results typically within ±20%. Speed depends on batch size, quantization kernel, and software stack.
What's the download size of Qwen3 4B Base?
At Q4_K_M, the download is about 2.41 GB. The full-precision BF16 version is 8.04 GB. The smallest option (IQ2_XXS) is 1.11 GB.
Which GPUs can run Qwen3 4B Base?
52 consumer GPUs can run Qwen3 4B Base at Q4_K_M. Top options include AMD Radeon RX 6700 XT, AMD Radeon RX 6800, AMD Radeon RX 6800 XT. 52 GPUs have plenty of headroom for comfortable inference.
Which devices can run Qwen3 4B Base?
163 devices with unified memory can run Qwen3 4B Base at Q4_K_M (2.9 GB), including AMD Ryzen AI 9 HX 370 (Strix Point) Laptop, ASUS Ascent GX10, Beelink GTR9 Pro (Ryzen AI Max+ 395, 128 GB), Framework Desktop (Ryzen AI Max+ 395, 128 GB). Apple Silicon Macs use unified memory shared between CPU and GPU, making them well-suited for local LLM inference.