What is Qwen3-235B-A22B FP8 good for?

Self-hosted reasoning and chat with switchable thinking and non-thinking modes. Apache 2.0 weights, Alibaba's reduced-memory FP8 build of the 235B mixture of experts, on a multi-GPU node.

When is Qwen3-235B-A22B FP8 not a good fit?

Image, video or audio input: text only, and its native context limits long-document work. Memory follows the full 235B parameters, not the 22B active, so it still takes a multi-GPU node.

What is the cheapest way to run Qwen3-235B-A22B FP8?

Hosted, unless you push serious volume. Alibaba Cloud charges $0.70 in / $2.80 out per 1M tokens. The cheapest rental that fits is 8x RTX 3090 at $979 a month, which costs the same as about 933M tokens a month on that API.

Can I self-host Qwen3-235B-A22B FP8?

Yes. The weights are Apache 2.0 licensed. At 4-bit it needs about 138 GB of GPU memory, which starts at roughly $979 a month on the cheapest rental that fits.

Open weights · Apache 2.0

Qwen3-235B-A22B FP8

FP8-quantized version of Qwen3-235B-A22B, a 235B MoE with 22B active, with a reduced memory footprint at near-identical quality.

Jump to: Specs Pricing Self-hosting Similar models FAQ More from Alibaba Cloud

Visit website Compare vs other LLMs →

List price via Alibaba Cloud

Input: $0.70 /1M tokens
Output: $2.80 /1M tokens

Cheapest: $0.20 / $0.80 per 1M tokens via Novita

2 providers

Key Specifications

Context window: 33K tokens
Parameters: 235B, 22B active
Inputs: Text

Capabilities Show details: Reasoning Thinks before it answers, always on or as a switchable mode. Function calling Connect to external tools, APIs, and systems. Structured output Return responses in structured formats like JSON.

Hosted API pricing

Input tokens M Output tokens M

Provider	Input /1M tokens	Output /1M tokens	Cost at 10M in + 2M out
Novita Cheapest	$0.20	$0.80	$3.60	View
Alibaba Cloud Creator	$0.70	$2.80	$12.60	View

Heads up: Base-tier, on-demand rates per 1M tokens; cached, batch and long-context tiers excluded. A provider may serve a shorter context or a quantized build than the creator's release. Verify before provisioning. More on how we price.

Estimated cost to self-host

Qwen3-235B-A22B FP8 needs about 138 GB of GPU memory at 4-bit with its full 32K context. The cheapest rental that fits is 8× RTX 3090 at about $979 a month.

That costs the same as roughly 933M tokens a month on Alibaba Cloud's API. Self-hosting is more expensive below that volume.

Single node ≤ 8

Precision	Memory	Cheapest fit (Full 32K context)	Cost /mo	Break-even vs API
4-bitINT4 / FP4	138 GB	8× RTX 3090	$979	933M tokens /mo
8-bitFP8 / INT8	243 GB	8× RTX A6000	$3,168	3B tokens /mo
16-bitFP16 / BF16	478 GB	8× A100	$10,138	10B tokens /mo

Estimates based on median on-demand rates for Nvidia GPUs. Memory is weights plus KV cache for one request, using FP8 KV cache where supported and FP16/BF16 otherwise. Break-even assumes a 5:1 input-to-output ratio. No guarantee of runtime support, usable performance, or that a matching quantized build exists. How we estimate costs.

Similarly priced models

The models nearest Qwen3-235B-A22B FP8 by blended rate, each at its own cheapest provider.

Model	Blended /1M	Input /1M	Output /1M	Context	Cutoff	vs Qwen3-235B-A22B FP8
Qwen3.5-122B-A10B Alibaba Cloud	$0.2533	$0.12	$0.92	262K		−16%
Llama 4 Maverick Meta	$0.2833	$0.20	$0.70	1M	Aug 2024	−6%
Qwen3-VL-30B-A3B Instruct Alibaba Cloud	$0.2833	$0.20	$0.70	262K		−6%
Muse Glimmer 30B Meta	$0.30	$0.20	$0.80	131K	Jan 2026	0%
Qwen2.5-7B Instruct Alibaba Cloud	$0.30	$0.30	$0.30	33K		0%
Qwen3-235B-A22B FP8 This model Alibaba Cloud	$0.30	$0.20	$0.80	33K
Qwen3.5-397B-A17B Alibaba Cloud	$0.3133	$0.17	$1.03	262K		+4%
MiniMax-M3 MiniMax	$0.3167	$0.20	$0.90	1M		+6%
Qwen3-Next-80B-A3B Instruct Alibaba Cloud	$0.325	$0.15	$1.20	262K		+8%
Qwen3-Next-80B-A3B Thinking Alibaba Cloud	$0.325	$0.15	$1.20	262K		+8%
Qwen-MT-Plus Alibaba Cloud	$0.3333	$0.25	$0.75	16K		+11%

Prices are USD per 1M tokens at each model's cheapest listed provider. Blended is the cost of 10M input plus 2M output tokens, spread over the 12M.

GPUhub

Fully Dedicated GPU Servers

No Virtualization. No Shared GPUs

Stable Performance for AI Training & Inference

Full Control for Production Workloads

Learn more

Frequently Asked Questions

What is Qwen3-235B-A22B FP8 good for?: Self-hosted reasoning and chat with switchable thinking and non-thinking modes. Apache 2.0 weights, Alibaba's reduced-memory FP8 build of the 235B mixture of experts, on a multi-GPU node.
When is Qwen3-235B-A22B FP8 not a good fit?: Image, video or audio input: text only, and its native context limits long-document work. Memory follows the full 235B parameters, not the 22B active, so it still takes a multi-GPU node.
What is the cheapest way to run Qwen3-235B-A22B FP8?: Hosted, unless you push serious volume. Alibaba Cloud charges $0.70 in / $2.80 out per 1M tokens. The cheapest rental that fits is 8x RTX 3090 at $979 a month, which costs the same as about 933M tokens a month on that API.
Can I self-host Qwen3-235B-A22B FP8?: Yes. The weights are Apache 2.0 licensed. At 4-bit it needs about 138 GB of GPU memory, which starts at roughly $979 a month on the cheapest rental that fits.

More from Alibaba Cloud

Model	Context	Input /1M	Output /1M
Qwen2.5-7B Instruct Open weights	33K	$0.30	$0.30
Qwen3.5-397B-A17B Open weights	262K	$0.17	$1.03
Qwen3-VL-30B-A3B Instruct Open weights	262K	$0.20	$0.70
Qwen3-Next-80B-A3B Instruct Open weights	262K	$0.15	$1.20
Qwen3-Next-80B-A3B Thinking Open weights	262K	$0.15	$1.20
Qwen2.5-VL-72B Instruct Open weights	33K	$0.25	$0.75
Qwen-MT-Plus	16K	$0.25	$0.75
Qwen3.6-27B Open weights	262K	$0.30	$0.60