LLMBOTTLENECK.COM
Qwen / qwen3_moe

Qwen3-235B-A22B-Instruct-2507

235.1 billion parameters, routing 8 of 128 experts per token.

Open in the calculator →

Architecturepublished data
Quick answer
Qwen3-235B-A22B-Instruct-2507 needs about 144.5 GB at Q4_K_M — the format most people download — with 8,192 tokens of context. It fits entirely on 1 of 18 common devices, starting with the Apple M2 Ultra (192 GB). What else fits in 192 GB. No hardware for it? Rent a GPU that runs it, priced live.

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 Apple M2 Ultra, 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.

Weight-file size by format

FormatSizeRangeBasis
FP16470.3 GB467.9 GB – 472.6 GBreconstructed size
Q8_0249.9 GB248.7 GB – 251.2 GBreconstructed size
Q6_K193.0 GB192.1 GB – 194.0 GBreconstructed size
Q5_K_M166.8 GB166.0 GB – 167.6 GBreconstructed size
Q5_0161.9 GB161.1 GB – 162.7 GBreconstructed size
Q4_K_M142.2 GB141.4 GB – 142.9 GBreconstructed size
Q4_0132.6 GB131.9 GB – 133.2 GBreconstructed size
Q3_K_M116.2 GB108.0 GB – 124.3 GBreconstructed size
Q2_K85.3 GB81.9 GB – 88.8 GBreconstructed size

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 · 1 of 18 devices hold it entirely
DeviceMemoryVerdictDecodeCalculator
NVIDIA GeForce RTX 5070144.5 GB of 12 GB88 layers on system RAM~4.8 tok/s with offloadOpen →
AMD Radeon™ RX 9070 XT144.5 GB of 16 GB85 layers on system RAM~4.9 tok/s with offloadOpen →
NVIDIA GeForce RTX 4080144.5 GB of 16 GB85 layers on system RAM~4.9 tok/s with offloadOpen →
NVIDIA GeForce RTX 5080144.5 GB of 16 GB85 layers on system RAM~5.0 tok/s with offloadOpen →
NVIDIA GeForce RTX 5070 Ti144.5 GB of 16 GB85 layers on system RAM~5.0 tok/s with offloadOpen →
NVIDIA GeForce RTX 5060 Ti 16GB144.5 GB of 16 GB85 layers on system RAM~4.8 tok/s with offloadOpen →
NVIDIA GeForce RTX 4090144.5 GB of 24 GB80 layers on system RAM~5.2 tok/s with offloadOpen →
AMD Radeon™ RX 7900 XTX144.5 GB of 24 GB80 layers on system RAM~5.3 tok/s with offloadOpen →
NVIDIA GeForce RTX 3090144.5 GB of 24 GB80 layers on system RAM~5.2 tok/s with offloadOpen →
NVIDIA GeForce RTX 5090144.5 GB of 32 GB75 layers on system RAM~5.6 tok/s with offloadOpen →
NVIDIA RTX 6000 Ada Generation144.5 GB of 48 GB64 layers on system RAM~6.4 tok/s with offloadOpen →
Apple M4 Pro144.5 GB of 64 GBdoes not fitnot calibratedOpen →
Apple M5 Pro144.5 GB of 64 GBdoes not fitnot calibratedOpen →
AMD Ryzen AI Max+ 395 with Radeon 8060S144.5 GB of 128 GBdoes not fitnot calibratedOpen →
Apple M3 Max144.5 GB of 128 GBdoes not fitnot calibratedOpen →
Apple M4 Max144.5 GB of 128 GBdoes not fitnot calibratedOpen →
NVIDIA DGX Spark144.5 GB of 128 GBdoes not fitnot calibratedOpen →
Apple M2 Ultra144.5 GB of 192 GBfits~40 tok/sOpen →

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 Qwen3-235B-A22B-Instruct-2507

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
10 configurations in stock · 10 out of stock
MachineSpeedPer hourPer M tokensWhereRent
4× L40S192 GBcheapest~160 tok/sFaster than you read$1.87$3.24Vast.aiMarketplace · 99.2% reliableRent on Vast.ai
2× RTX PRO 6000192 GB~166 tok/sFaster than you read$2.27$3.80Vast.aiMarketplace · 98.8% reliableRent on Vast.ai
4× RTX 6000 Ada192 GB~177 tok/sFaster than you read$3.00$4.68Vast.aiMarketplace · 98.5% reliableRent on Vast.ai
4× RTX PRO 5000192 GB~249 tok/sFaster than you read$3.28$3.66RunPodCommunity CloudRent on RunPod
2× H100 SXM160 GBrecommendedbest value~310 tok/sFaster than you read$3.47$3.11Vast.aiMarketplace · 98.7% reliableRunPod $6.98/h Rent on Vast.ai
8× RTX 4090192 GB~337 tok/sFaster than you read$4.06$3.34Vast.aiMarketplace · 99.2% reliableRent on Vast.ai
2× H100 PCIe160 GB~185 tok/sFaster than you read$4.79$7.18Vast.aiMarketplace · 98.7% reliableRent on Vast.ai
2× H200 NVL282 GB~444 tok/sFaster than you read$7.20$4.50Vast.aiMarketplace · 98.7% reliableRent on Vast.ai

No GPU at all

Use Qwen3-235B-A22B-Instruct-2507 by the token

For one person chatting, that is about 8.9× cheaper than the best-value rented card above ($3.11 per million tokens). A rented GPU pays off when you keep it busy — many requests batched together, long agent runs — or when the data must stay on a machine you control, or the exact file you want is not served anywhere.

First time renting a GPU? How it works, in four steps
  1. 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.
  2. 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.
  3. 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.
  4. 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 Qwen3-235B-A22B-Instruct-2507 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 Qwen3-235B-A22B-Instruct-2507 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 read from the published config.json

Retrieved 2026-09-01 at pinned commit ac9c66cc9b46; parameter count from the safetensors index at the same revision.

Architecture
qwen3_moe
Layers
94
Hidden size
4,096
Attention heads
64
KV heads
4
Head dimension
128
Feed-forward width
12,288
Vocabulary
151,936
Context ceiling
262,144
RoPE theta
5,000,000
Experts
128
Experts / token
8
Expert width
1,536

Where the memory goes

tokenembedding× 94 decoder blocksGrouped-query attention64 query · 4 KV headsfull contextRouted experts8 of 128 per tokenall residentoutputprojectiongrows with contextcapacity ≠ traffic

Grouped-query attention shares each key/value head across 16 query heads, so the KV cache is 6% of what multi-head attention would need at the same context.

Open the exact official config.json ↗

More from Qwen

Qwen-AgentWorld-35B-A3BQwen2.5-0.5BQwen2.5-0.5B-InstructQwen2.5-1.5B-InstructQwen2.5-14B-InstructQwen2.5-32B-InstructQwen2.5-3B-InstructQwen2.5-72B-Instruct

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-qwen3-235b-a22b-instruct-2507","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-qwen3-235b-a22b-instruct-2507" 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. “Qwen3-235B-A22B-Instruct-2507 VRAM and hardware requirements.” Architecture from Qwen/Qwen3-235B-A22B-Instruct-2507 at revision ac9c66cc9b46, retrieved 2026-09-01. https://llmbottleneck.com/models/qwen-qwen3-235b-a22b-instruct-2507

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

llmbottleneck
catalogue 2026-10-03models 327devices 135