Qwen / qwen3_vl_text

Qwen3-VL-2B-Instruct

Name: Qwen3-VL-2B-Instruct architecture and memory requirements
Creator: LLM Bottleneck
License: https://creativecommons.org/licenses/by/4.0/

2.13 billion parameters.

Open in the calculator →

Architecturepublished data

Quick answer

Qwen3-VL-2B-Instruct needs about 3.05 GB at Q4_K_M — the format most people download — with 8,192 tokens of context. It fits entirely on 18 of 18 common devices, starting with the NVIDIA GeForce RTX 5070 (12 GB). What else fits in 12 GB. No hardware for it? Rent a GPU that runs it, priced live.

Smallest device that fitsNVIDIA GeForce RTX 507012 GB · 672 GB/s~224 tok/s Faster than you readOpen in the calculator →
Fastest in this listNVIDIA GeForce RTX 509032 GB · 1792 GB/s~596 tok/s Faster than you readOpen in the calculator →

Sized at 8,192 tokens of context with the whole model in device memory, across 18 common devices. Speeds are estimates, not benchmarks.

Find the NVIDIA GeForce RTX 5070, the smallest card listed here that holds it: Amazon ↗ · eBay (new and used) ↗Store links may pay us a commission. They never decide which card is suggested — the memory arithmetic does.

Vision-language model. Sized for its text decoder; image inputs add memory not modelled here. Licence declared on the repository: apache-2.0.

Weight-file size by format

every format, with its evidence

Format	Size	Range	Basis
FP16	4.26 GB	2.98 GB – 4.30 GB	size range
Q8_0	2.26 GB	1.58 GB – 2.87 GB	size range
Q6_K	1.75 GB	1.22 GB – 2.50 GB	size range
Q5_K_M	1.52 GB	1.02 GB – 2.50 GB	size range
Q5_0	1.48 GB	1.02 GB – 2.50 GB	size range
Q4_K_M	1.31 GB	0.84 GB – 2.50 GB	size range
Q4_0	1.24 GB	0.84 GB – 2.50 GB	size range
Q3_K_M	1.06 GB	0.64 GB – 2.50 GB	size range
Q2_K	0.84 GB	0.49 GB – 2.50 GB	size range

A file size is not the memory a run needs: the KV cache and the runtime reserve come on top, and the calculator adds both.

What the labels mean

published data: Read from a published source — a file's byte count, a model's configuration or a manufacturer's specification — or exact arithmetic on such values. Not a measurement on a machine.
reconstructed size: Weight size reconstructed from the pinned architecture, because no published file exists.
size range: Only a lower and an upper bound are claimed for this weight size.

Hardware ladder

Q4_K_M at 8,192 tokens · 18 of 18 devices hold it entirely

Device	Memory	Verdict	Decode	Calculator
NVIDIA GeForce RTX 5070	3.05 GB of 12 GB	fits	~224 tok/s	Open →
AMD Radeon™ RX 9070 XT	3.05 GB of 16 GB	fits	~244 tok/s	Open →
NVIDIA GeForce RTX 4080	3.05 GB of 16 GB	fits	~239 tok/s	Open →
NVIDIA GeForce RTX 5080	3.05 GB of 16 GB	fits	~320 tok/s	Open →
NVIDIA GeForce RTX 5070 Ti	3.05 GB of 16 GB	fits	~298 tok/s	Open →
NVIDIA GeForce RTX 5060 Ti 16GB	3.05 GB of 16 GB	fits	~149 tok/s	Open →
NVIDIA GeForce RTX 4090	3.05 GB of 24 GB	fits	~335 tok/s	Open →
AMD Radeon™ RX 7900 XTX	3.05 GB of 24 GB	fits	~366 tok/s	Open →
NVIDIA GeForce RTX 3090	3.05 GB of 24 GB	fits	~312 tok/s	Open →
NVIDIA GeForce RTX 5090	3.05 GB of 32 GB	fits	~596 tok/s	Open →
NVIDIA RTX 6000 Ada Generation	3.05 GB of 48 GB	fits	~320 tok/s	Open →
Apple M4 Pro	3.05 GB of 64 GB	fits	~73 tok/s	Open →
Apple M5 Pro	3.05 GB of 64 GB	fits	~78 tok/s	Open →
AMD Ryzen AI Max+ 395 with Radeon 8060S	3.05 GB of 128 GB	fits	~98 tok/s	Open →
Apple M3 Max	3.05 GB of 128 GB	fits	~90 tok/s	Open →
Apple M4 Max	3.05 GB of 128 GB	fits	~103 tok/s	Open →
NVIDIA DGX Spark	3.05 GB of 128 GB	fits	~91 tok/s	Open →
Apple M2 Ultra	3.05 GB of 192 GB	fits	~119 tok/s	Open →

Decode is a calibrated estimate from the published calibration; capacity is the manufacturer's published ceiling, not guaranteed free memory. Fitting in memory is not the same as loading: whether the runtime and version you have supports this architecture and format on that machine has not been tested here. Offloaded rows assume there is enough system RAM for the overflow — the calculator checks that against the RAM you declare. Each “Open” link keeps this model, format and context.

Run it in the cloud

Rent a GPU that runs Qwen3-VL-2B-Instruct

Every rentable machine that holds the whole model, from Vast.ai and RunPod, with this site's speed estimate for each and the provider's own price, read live. Cheapest first; sort by cost per token or speed instead, or switch the format.

Prices · 18:30 UTC, 22 Sept

Machine	Speed	Per hour	Per M tokens	Where	Rent
1× RTX 306012 GBrecommendedcheapestbest value	~119 tok/sFaster than you read	$0.036	$0.083	Vast.aiMarketplace · 99.8% reliable	Rent on Vast.ai
1× RTX 308010 GB	~252 tok/sFaster than you read	$0.082	$0.090	Vast.aiMarketplace · 98.3% reliable	Rent on Vast.ai
1× RTX 30708 GB	~149 tok/sFaster than you read	$0.082	$0.15	Vast.aiMarketplace · 99.1% reliableRunPod $0.13/h	Rent on Vast.ai
1× RTX 4060 Ti 16GB16 GB	~95 tok/sFaster than you read	$0.090	$0.26	Vast.aiMarketplace · 99.4% reliable	Rent on Vast.ai
1× RTX 407012 GB	~167 tok/sFaster than you read	$0.096	$0.16	Vast.aiMarketplace · 99.8% reliable	Rent on Vast.ai
1× RTX 309024 GB	~311 tok/sFaster than you read	$0.12	$0.11	Vast.aiMarketplace · 99.5% reliableRunPod $0.50/h	Rent on Vast.ai
1× RTX 5060 Ti 16GB16 GB	~149 tok/sFaster than you read	$0.12	$0.23	Vast.aiMarketplace · 99.1% reliable	Rent on Vast.ai
1× RTX 3080 Ti12 GB	~303 tok/sFaster than you read	$0.14	$0.12	Vast.aiMarketplace · 99.9% reliable	Rent on Vast.ai

First time renting a GPU? How it works, in four steps

Create an account and add credit. Both providers are prepaid: $10 is enough to try any card on this page for hours. A new RunPod account that arrives through a referral link gets a one-time $5 credit when it first adds $10.
Pick the card and the count shown here, and a template: llama.cpp or vLLM for an OpenAI-compatible API, or one with a web chat if you only want to talk to the model. “How to launch it” below gives the exact image and command.
Wait for the download. The weights are fetched on the machine; tens of gigabytes take a few minutes on a data-centre connection. You pay by the second from the moment it starts.
Stop it when you are done. A running machine bills even when idle. A stopped one costs nothing per hour, though its disk is usually still billed until you delete it.

How to launch it

No Q4_K_M file of Qwen3-VL-2B-Instruct is catalogued here, so there is no exact command to vouch for. Start the machine from the llama.cpp server image ghcr.io/ggml-org/llama.cpp:server-cuda and point -hf at a Q4_K_M build from the publisher or a quantizer you trust — search Hugging Face for one. Check its file size against the memory figure on this page before you rent.

Every machine holds the whole model at Q4_K_M and 8,192 tokens of context, no offload. Speeds are this site’s single-stream decode estimates; prices are what each provider’s own API quoted, on-demand, for the whole machine. Vast hosts below 98% measured reliability are left out.

Referral links Vast.ai, RunPod and Novita pay us a share of what you spend if you sign up through these buttons. It costs you nothing, and it never decides an order or a recommendation: both are computed from the live price and the speed, and options that pay us nothing are listed and recommended on the same terms. How we rank

Buying a card instead, or paying by the token? Compare a year of Qwen3-VL-2B-Instruct three ways →

Under the hood

Architecture, read from the publisher’s file

The numbers every figure above is computed from, with the file they came from.

Architecture

architecture.json · pinned 89644892e4d8

✓ Architecture read from the published config.json

Retrieved 2026-09-13 at pinned commit 89644892e4d8; parameter count from the safetensors index at the same revision.

Architecture: qwen3_vl_text
Layers: 28
Hidden size: 2,048
Attention heads: 16
KV heads: 8
Head dimension: 128
Feed-forward width: 6,144
Vocabulary: 151,936
Context ceiling: 262,144
RoPE theta: 5,000,000

Where the memory goes

Grouped-query attention shares each key/value head across 2 query heads, so the KV cache is 50% of what multi-head attention would need at the same context.

This is a multimodal checkpoint (vision); the diagram and memory sizing above cover the text decoder. Encoder/audio/image activations are not included in the KV or weight figures.

Open the exact official config.json ↗

Qwen3-VL-2B-Instruct, card by card

One page per device: whether Qwen3-VL-2B-Instruct fits, at which formats, and how fast.

All 27Fewer

Derivatives this page also answers for

Each of these repositories declares Qwen3-VL-2B-Instruct as its base, and its published configuration matches this one on every field that decides memory. Its page lists what was compared and any files it publishes. Paste any other repository.

typhoon-ai/typhoon-ocr1.5-2b

Use this answer in your own product

As JSON, from the API

curl -s https://llmbottleneck.com/v1/analyze \
  -H "Authorization: Bearer $LLMB_KEY" -H "content-type: application/json" \
  -d '{"model":"qwen-qwen3-vl-2b-instruct","quantization":"Q4_K_M","context":8192,"hardware":"nvidia-geforce-rtx-4090"}'

Same engine, same evidence, every field sourced. Free key in one step →

On your page, as a widget

<script src="https://llmbottleneck.com/widget.js"
  data-model="qwen-qwen3-vl-2b-instruct" data-quantization="Q4_K_M"
  data-hardware="nvidia-geforce-rtx-4090" data-context="8192"></script>

No key needed. Unbranded, with your own buy button, on Pro and Business →

Citing this page

LLM Bottleneck. “Qwen3-VL-2B-Instruct VRAM and hardware requirements.” Architecture from Qwen/Qwen3-VL-2B-Instruct at revision 89644892e4d8, retrieved 2026-09-13. https://llmbottleneck.com/models/qwen-qwen3-vl-2b-instruct

Every figure above is either the published value or a reconstruction whose measured error is on the accuracy page.