Qwen3.5 2B Base — Hardware Requirements & GPU Compatibility
Vision
Qwen3.5-2B-Base is a 2.3-billion-parameter dense checkpoint in Alibaba's Qwen3.5 base model family, a native vision-language foundation model rather than a text-only model with vision bolted on. It is pretrain-only: fine-tuning, in-context-learning, or further research, not direct conversation, though its control tokens are compatible with the official chat template for efficient LoRA-style adaptation. Its hybrid architecture pairs Gated DeltaNet linear attention with periodic gated full-attention layers. At 2.3B parameters, it runs comfortably on a single consumer GPU, even unquantized. The model supports a native 262,144 token context window, extensible up to 1,010,000 tokens. It is released under the Apache 2.0 license, and was published in February 2026. Qwen3.5 introduced early-fusion multimodal pretraining that Alibaba says outperforms the separate Qwen3-VL models on reasoning, coding, and visual understanding.
Runs great on a 24 GB GPU
Runs from 8 GB (Q4_K_M · 256K)Best on 24 GB (Q4_K_M · ~370 tok/s · 256K)Macs from 8 GB (Q4_K_M · ~41 tok/s)
Apache 2.0
Reference GPU · RTX 4090 24 GB
Estimates, ±20% · how we estimate
Run Qwen3.5 2B Base
One recommended way. Every other method, OS and quant is in the step-by-step guide.Each way opens a step-by-step guide for your OS and quant.
Recommended
ollama run hf.co/mradermacher/Qwen3.5-2B-Base-i1-GGUF:Q4_K_MQ4_K_M fits fully on the RTX 4090 · 22.1 GB free · 16K context
Download: Qwen3.5-2B-Base.i1-Q4_K_M.gguf1.27 GBmradermacher
Ways to runOther ways
each opens step-by-stepRun in cloud
Runs on RTX 3080 Ti (12 GB) (10 GB headroom) · Q4_K_M
- Generation speed
- ~335 tok/s
- generation speed
- Cost per 1M output tokens
- $0.10
- per 1M output tokens
How Much VRAM Does Qwen3.5 2B Base Need?
Select a quantization level; the listed hardware adheres to it.
| Quantization | VRAM | +32K | On RTX 4090 (reference) | |
|---|---|---|---|---|
| 1.4 GB | 1.7 GB | 2-bit · K-quant — 2-bit quantization with K-quant improvements | Fits: 256K context · ~478 tok/s | |
| 1.4 GB | 1.7 GB | 3-bit small — 3-bit small quantization | Fits: 256K context · ~468 tok/s | |
| 1.5 GB | 1.8 GB | 3-bit medium — 3-bit medium quantization | Fits: 256K context · ~434 tok/s | |
| 1.5 GB | 1.8 GB | 4-bit · legacy — 4-bit legacy quantization | Fits: 256K context · ~426 tok/s | |
| 1.8 GB | 2.1 GB | 4-bit medium — 4-bit medium quantization — most popular sweet spot | Fits: 256K context · ~370 tok/s | |
| 2.0 GB | 2.3 GB | 5-bit medium — 5-bit medium quantization — good quality/size tradeoff | Fits: 256K context · ~324 tok/s | |
| 2.3 GB | 2.6 GB | 6-bit, very good — 6-bit quantization, very good quality | Fits: 256K context · ~287 tok/s | |
| 2.7 GB | 3 GB | 8-bit, near-lossless — 8-bit quantization, near-lossless | Fits: 256K context · ~245 tok/s |
Selected Q4_K_M
4-bit medium quantization1.9GBat 16K context
- Weights
- 1.4 GB
- KV cache
- 0.2 GB
- Overhead
- 0.3 GB
Fits RTX 4090 (reference) · 22.1 GB free · ~370 tok/s
≥50 tok/s needs ≥136 GB/s · RTX 5090 ~658, RTX 3090 Ti ~370
Which Hardware Can Run Qwen3.5 2B Base?
Q4_K_M · 1.8 GBQwen3.5 2B Base (Q4_K_M) requires 1.8 GB of VRAM to load the model weights. For comfortable inference with headroom for KV cache and system overhead, 3+ GB is recommended. Using the full 262K context window can add up to 3.1 GB, bringing total usage to 4.9 GB. 52 GPUs can run it, including NVIDIA GeForce RTX 5090, NVIDIA GeForce RTX 3090 Ti. 163 devices with unified memory can run Qwen3.5 2B Base, including Mac Studio (M3 Ultra, 256GB), Mac Studio (M3 Ultra, 512GB).
≤ 11 GB
≤ 11 GB tier →fits · 6.1 GB free · up to 256K ctx
16 GPUs · 1 Mac class · 2 device families
RTX 3080~279Simulate output on RTX 3080RTX 2080 Ti~226Simulate output on RTX 2080 TiRTX 3070 Ti~223Simulate output on RTX 3070 TiM3*8 GB~41Simulate output on iMac 24" M3 (8 GB)Orin Nano~38Simulate output on NVIDIA Jetson Orin Nano 8GB (Super)iPhone~27Simulate output on iPhone 1720 GB
fits · 18.1 GB free · up to 256K ctx
1 GPU
RX 7900 XT~271Simulate output on RX 7900 XT32–40 GB
fits · 30.1 GB free · up to 256K ctx
1 GPU · 3 Mac classes · 4 device families
RTX 5090~658Simulate output on RTX 5090M5 Max36 GB~182Simulate output on Mac Studio M5 Max (36 GB)M532 GB~61Simulate output on MacBook Air 13" M5 (32 GB)Orin~75Simulate output on NVIDIA Jetson AGX Orin 32GBSnapdragon X Elite (X1E-80-100)~50Simulate output on Snapdragon X Elite Copilot+ PC
By VRAM tier · typical GPUs and Macs at Q4_K_M
Alternatives to Qwen3.5 2B Base
Qwen 3.5 family · 34 models →Similar models
6| Model | Score | VRAM | tok/s | Why it is similar |
|---|---|---|---|---|
| WeVisOne VL 2BTencent | — | 3.6 GB | ~328 | same size |
| LFM3 Omni 2BLiquid AI | — | 2.4 GB | ~385 | same size |
| SmolVLM2 2.2B InstructHugging Face | — | 2.3 GB | ~443 | same size |
| SmolVLM InstructHugging Face | — | 4.9 GB | ~320 | same size |
| InternVL2 2BOpenGVLab | — | 3.3 GB | ~449 | same size |
| Kimi K3 DSparkRadixArk | — | 2.2 GB | ~381 | same size |
Fine-tunes & variants(2)
Specifications
- Publisher
- Alibaba
- Family
- Qwen 3.5
- Parameters
- 2.3B
- Architecture
- Qwen3_5ForConditionalGeneration
- Context Length
- 262,144 tokens
- Vocabulary Size
- 248,320
- Release Date
- 2026-02-28
- License
- Apache 2.0
Frequently Asked Questions
How much VRAM does Qwen3.5 2B Base need?
Qwen3.5 2B Base requires 1.8 GB of VRAM at Q4_K_M, or 5.0 GB at BF16. Full 262K context adds up to 3.1 GB (4.9 GB total).
VRAM = Weights + KV Cache + Overhead
Weights = 2.3B × 4.8 bits ÷ 8 = 1.4 GB
KV Cache + Overhead ≈ 0.4 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 ≈ 3.5 GB (at full 262K context)
VRAM usage by quantization
What's the best quantization for Qwen3.5 2B Base?
For Qwen3.5 2B Base, Q4_K_M (1.8 GB) offers the best balance of quality and VRAM usage. Q5_0 (1.8 GB) provides better quality if you have the VRAM. The smallest option is IQ2_XXS at 1.0 GB.
VRAM requirement by quantization
★ Recommended — best balance of quality and VRAM usage.
Can I run Qwen3.5 2B Base on a Mac?
Yes — MacBook Air 13" M3 (8 GB) and 145 other Macs can run Qwen3.5 2B Base. Apple Silicon uses unified memory, so the model shares RAM with the system. At Q4_K_M you need at least 1.8 GB of usable unified memory (RAM minus macOS overhead).
Can I run Qwen3.5 2B Base locally?
Yes — Qwen3.5 2B Base can run locally on consumer hardware. Popular tools include Ollama, LM Studio, and llama.cpp.
How fast is Qwen3.5 2B Base?
At Q4_K_M, Qwen3.5 2B Base can reach ~2712 tok/s on AMD Instinct MI350X. On NVIDIA GeForce RTX 4090: ~370 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 ÷ 1.77 × 0.65 = ~2938 tok/s
Estimated speed at Q4_K_M (1.8 GB)
Real-world results typically within ±20%. Speed depends on batch size, quantization kernel, and software stack.
What's the download size of Qwen3.5 2B Base?
At Q4_K_M, the download is about 1.36 GB. The full-precision BF16 version is 4.55 GB. The smallest option (IQ2_XXS) is 0.63 GB.
Which GPUs can run Qwen3.5 2B Base?
52 consumer GPUs can run Qwen3.5 2B 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 Qwen3.5 2B Base?
163 devices with unified memory can run Qwen3.5 2B Base at Q4_K_M (1.8 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.