# Qwen3-235B-A22B GPU requirements

Qwen3-235B-A22B has 235B parameters. To serve it, plan for about 564 GB of GPU memory in 16-bit, 282 GB in 8-bit or 141 GB in 4-bit, including about 20% for the KV cache and activations.

It is a mixture of experts: 128 experts, 8 of them active for each token. All of them stay in GPU memory, so plan memory for all 235B parameters.

[Deploy it on Arrakis](/console/#/models?model=Qwen/Qwen3-235B-A22B).

## What it takes to serve it

| Precision | Memory and the smallest setup |
|---|---|
| 16-bit | 564 GB: 8 × [H100](/gpus/h100.html.md), from $13.20 per hour for all 8 |
| 8-bit | 282 GB: 4 × [H100](/gpus/h100.html.md), from $6.60 per hour for all 4 |
| 4-bit | 141 GB: one [B200](/gpus/b200.html.md), from $6.79 per GPU-hour |

## What it takes to fine-tune it

| Method | Memory and the smallest setup |
|---|---|
| QLoRA | 165 GB: one [B200](/gpus/b200.html.md), from $6.79 per GPU-hour |
| LoRA | 564 GB: 8 × [H100](/gpus/h100.html.md), from $13.20 per hour for all 8 |
| Full fine-tuning | 3,761 GB: more than one 8-GPU machine |

## About the model

| | |
|---|---|
| Parameters | `235B` |
| Active for each token | `22B` |
| Experts | `128` |
| Experts for each token | `8` |
| Context window (tokens) | `40,960` |
| Published in | `BF16` |
| License | `apache-2.0` |
| Released | `2025-04-27` |
| Weights | [Qwen/Qwen3-235B-A22B](https://huggingface.co/Qwen/Qwen3-235B-A22B) |

## Serve it yourself

```
vllm serve Qwen/Qwen3-235B-A22B --tensor-parallel-size 8
```

## Questions

**How much GPU memory does Qwen3-235B-A22B need?** To serve it, about 564 GB in 16-bit, 282 GB in 8-bit, 141 GB in 4-bit, including about 20% for the KV cache and activations. To fine-tune it, about 165 GB with QLoRA, 564 GB with LoRA and 3,761 GB for full fine-tuning with Adam, before activations.

**Can I run Qwen3-235B-A22B on one GPU?** Yes, quantized. In 4-bit it needs about 141 GB, which fits on one B200. In 16-bit it needs 8 H100 GPUs.

**What does it cost to run Qwen3-235B-A22B?** In 16-bit it needs 8 H100 GPUs; in 4-bit, one B200. Each setup on this page shows the lowest price per GPU-hour on Arrakis right now, and you pay by the hour from a prepaid balance.

**How do I deploy Qwen3-235B-A22B on Arrakis?** Open Models in the Arrakis console, paste Qwen/Qwen3-235B-A22B, choose vLLM, SGLang or llama.cpp, and rent a GPU it fits on. You get an OpenAI-compatible endpoint with its own API key, and root SSH to the machine.


More models: [/models](/models.html.md).