LLMBOTTLENECK.COM
Qwen / qwen3_next

Qwen3-Next-80B-A3B-Thinking

81.3 billion parameters, routing 10 of 512 experts per token.

Open in the calculator →

Architecturepublished data
Quick answer
Qwen3-Next-80B-A3B-Thinking needs about 50.0 GB at Q4_K_M — the format most people download — with 8,192 tokens of context. It fits entirely on 7 of 18 common devices, starting with the Apple M4 Pro (64 GB). What else fits in 64 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 M4 Pro, 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
FP16162.7 GB113.9 GB – 164.3 GBsize range
Q8_084.8 GBexactpublished data
Q6_K65.5 GBexactpublished data
Q5_K_M56.7 GBexactpublished data
Q5_055.0 GBexactpublished data
Q4_K_M48.4 GBexactpublished data
Q4_047.3 GB32.0 GB – 68.1 GBsize range
Q3_K_M40.7 GB24.5 GB – 68.1 GBsize range
Q2_K32.2 GB18.7 GB – 68.1 GBsize range

5 of these are published files. Open the pinned GGUF repository ↗

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 · 7 of 18 devices hold it entirely
DeviceMemoryVerdictDecodeCalculator
NVIDIA GeForce RTX 507050.0 GB of 12 GB38 layers on system RAM~35 tok/s with offloadOpen →
AMD Radeon™ RX 9070 XT50.0 GB of 16 GB34 layers on system RAM~38 tok/s with offloadOpen →
NVIDIA GeForce RTX 408050.0 GB of 16 GB34 layers on system RAM~38 tok/s with offloadOpen →
NVIDIA GeForce RTX 508050.0 GB of 16 GB34 layers on system RAM~39 tok/s with offloadOpen →
NVIDIA GeForce RTX 5070 Ti50.0 GB of 16 GB34 layers on system RAM~39 tok/s with offloadOpen →
NVIDIA GeForce RTX 5060 Ti 16GB50.0 GB of 16 GB34 layers on system RAM~36 tok/s with offloadOpen →
NVIDIA GeForce RTX 409050.0 GB of 24 GB26 layers on system RAM~49 tok/s with offloadOpen →
AMD Radeon™ RX 7900 XTX50.0 GB of 24 GB26 layers on system RAM~50 tok/s with offloadOpen →
NVIDIA GeForce RTX 309050.0 GB of 24 GB26 layers on system RAM~49 tok/s with offloadOpen →
NVIDIA GeForce RTX 509050.0 GB of 32 GB18 layers on system RAM~71 tok/s with offloadOpen →
NVIDIA RTX 6000 Ada Generation50.0 GB of 48 GB2 layers on system RAM~181 tok/s with offloadOpen →
Apple M4 Pro50.0 GB of 64 GBfits~60 tok/sOpen →
Apple M5 Pro50.0 GB of 64 GBfits~65 tok/sOpen →
AMD Ryzen AI Max+ 395 with Radeon 8060S50.0 GB of 128 GBfits~72 tok/sOpen →
Apple M3 Max50.0 GB of 128 GBfits~76 tok/sOpen →
Apple M4 Max50.0 GB of 128 GBfits~90 tok/sOpen →
NVIDIA DGX Spark50.0 GB of 128 GBfits~67 tok/sOpen →
Apple M2 Ultra50.0 GB of 192 GBfits~106 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-Next-80B-A3B-Thinking

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
19 configurations in stock · 14 out of stock
MachineSpeedPer hourPer M tokensWhereRent
4× RTX 5060 Ti 16GB64 GBcheapest~339 tok/sFaster than you read$0.51$0.42Vast.aiMarketplace · 99.5% reliableRent on Vast.ai
2× RTX 509064 GBrecommendedbest value~874 tok/sFaster than you read$0.54$0.17Vast.aiMarketplace · 99.6% reliableRent on Vast.ai
8× RTX 306096 GB~376 tok/sFaster than you read$0.59$0.43Vast.aiMarketplace · 99.7% reliableRent on Vast.ai
4× RTX 309096 GB~708 tok/sFaster than you read$0.60$0.23Vast.aiMarketplace · 99.4% reliableRent on Vast.ai
4× RTX 4060 Ti 16GB64 GB~218 tok/sFaster than you read$0.80$1.02Vast.aiMarketplace · 99.3% reliableRent on Vast.ai
2× RTX 6000 Ada96 GB~468 tok/sFaster than you read$1.07$0.64Vast.aiMarketplace · 99.4% reliableRent on Vast.ai
1× RTX PRO 600096 GB~437 tok/sFaster than you read$1.14$0.72Vast.aiMarketplace · 98.8% reliableRunPod $1.69/h Rent on Vast.ai
1× A100 80GB PCIe80 GB~472 tok/sFaster than you read$1.19$0.70RunPodCommunity CloudRent on RunPod

No GPU at all

Use Qwen3-Next-80B-A3B-Thinking by the token

Here the rented card is as cheap as the API or cheaper per token ($0.17 per million), and it keeps your data on a machine you control. The API is still the easier start: nothing to set up and nothing to switch off.

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 Q4_K_M on it (llama.cpp)
Container image
ghcr.io/ggml-org/llama.cpp:server-cuda
Start command / arguments
-hf Qwen/Qwen3-Next-80B-A3B-Thinking-GGUF:Q4_K_M -c 8192 -ngl 999 --host 0.0.0.0 --port 8080
Expose port
8080 — an OpenAI-compatible API at /v1

Starting from a saved llama.cpp template instead? Leave the arguments empty and set these variables — only the first two change between models:

LLAMA_ARG_HF_REPO=Qwen/Qwen3-Next-80B-A3B-Thinking-GGUF:Q4_K_M
LLAMA_ARG_CTX_SIZE=8192
LLAMA_ARG_N_GPU_LAYERS=999
LLAMA_ARG_HOST=0.0.0.0
LLAMA_ARG_PORT=8080

Already have the runtime? llama-server -hf Qwen/Qwen3-Next-80B-A3B-Thinking-GGUF:Q4_K_M -c 8192 -ngl 999 --host 0.0.0.0 --port 8080. Weights: Qwen/Qwen3-Next-80B-A3B-Thinking-GGUF. The CUDA image is for NVIDIA cards; an AMD machine needs llama.cpp's ROCm build.

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-Next-80B-A3B-Thinking 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 e502dd4100cc; parameter count from the safetensors index at the same revision.

Architecture
qwen3_next
Layers
48
Hidden size
2,048
Attention heads
16
KV heads
2
Head dimension
256
Feed-forward width
5,120
Vocabulary
151,936
Context ceiling
262,144
RoPE theta
10,000,000
Experts
512
Experts / token
10
Expert width
512
Shared expert width
512

Where the memory goes

tokenembedding× 48 decoder blocksGrouped-query attention16 query · 2 KV headsfull contextRouted experts10 of 512 per tokenall residentoutputprojectiongrows with contextcapacity ≠ traffic

Grouped-query attention shares each key/value head across 8 query heads, so the KV cache is 13% 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-next-80b-a3b-thinking","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-next-80b-a3b-thinking" 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-Next-80B-A3B-Thinking VRAM and hardware requirements.” Architecture from Qwen/Qwen3-Next-80B-A3B-Thinking at revision e502dd4100cc, retrieved 2026-09-01. https://llmbottleneck.com/models/qwen-qwen3-next-80b-a3b-thinking

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