What is Qwen3-32B FP8 good for?

Self-hosted reasoning and chat with a switch between thinking and non-thinking modes, covering over 100 languages, Alibaba says. Apache 2.0 weights, on a single high-memory GPU.

When is Qwen3-32B FP8 not a good fit?

Image, video or audio input: text only, with a short native context. It runs all 32B parameters on every token, unlike the 30B-A3B mixture of experts with 3B active, so each token takes more compute to serve.

What is the cheapest way to run Qwen3-32B FP8?

Hosted, unless you push serious volume. Alibaba Cloud charges $0.16 in / $0.64 out per 1M tokens. The cheapest rental that fits is 2x RTX 4060 Ti at $158 a month, which costs the same as about 660M tokens a month on that API.

Can I self-host Qwen3-32B FP8?

Yes. The weights are Apache 2.0 licensed. At 4-bit it needs about 24 GB of GPU memory, which starts at roughly $158 a month on the cheapest rental that fits.

Open weights · Apache 2.0

Qwen3-32B FP8

FP8-quantized version of the 32B dense Qwen3 model, with switching between reasoning and general dialogue modes.

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: $0.16 /1M tokens
Output: $0.64 /1M tokens

Cheapest: $0.10 / $0.30 per 1M tokens via Lyceum

2 providers

Key Specifications

Context window: 33K tokens
Parameters: 32.8B
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
Lyceum Cheapest	$0.10	$0.30	$1.60	View
Alibaba Cloud Creator	$0.16	$0.64	$2.88	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

Qwen3-32B FP8 needs about 24 GB of GPU memory at 4-bit with its full 32K context. The cheapest rental that fits is 2× RTX 4060 Ti at about $158 a month.

That costs the same as roughly 660M 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	24 GB	2× RTX 4060 Ti	$158	660M tokens /mo
8-bitFP8 / INT8	39 GB	4× RTX 4070	$288	1B tokens /mo
16-bitFP16 / BF16	76 GB	8× RTX 3060	$346	1B 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 Qwen3-32B FP8 by blended rate, each at its own cheapest provider.

Model	Blended /1M	Input /1M	Output /1M	Context	Cutoff	vs Qwen3-32B FP8
GPT-5 Nano OpenAI	$0.1083	$0.05	$0.40	400K	May 2024	−19%
Qwen3.5-35B-A3B Alibaba Cloud	$0.1267	$0.06	$0.46	262K		−5%
Qwen3.6-35B-A3B Alibaba Cloud	$0.1283	$0.07	$0.42	262K		−4%
Gemma 3 27B Instruct Google Cloud	$0.1333	$0.10	$0.30	128K	Aug 2024	0%
Gemma 4 26B A4B Google Cloud	$0.1333	$0.10	$0.30	256K	Jan 2025	0%
Qwen3-32B FP8 This model Alibaba Cloud	$0.1333	$0.10	$0.30	33K
Gemini 2.5 Flash-Lite Google Cloud	$0.15	$0.10	$0.40	1M	Jan 2025	+13%
GPT-4.1 Nano OpenAI	$0.15	$0.10	$0.40	1M	Jun 2024	+13%
Ministral 3 8B Mistral	$0.15	$0.15	$0.15	256K		+13%
Qwen3.5-Flash Alibaba Cloud	$0.15	$0.10	$0.40	1M		+13%
Llama 3.2 11B Vision Meta	$0.155	$0.05	$0.68	128K	Dec 2023	+16%

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.

Lyceum

NVIDIA H200 from $2.20/hr, B200 from $3.40/hr, B300 from $3.80/hr

Per-second billing with no minimum commitment or vendor lock-in

Scale to 8,000 GPUs over InfiniBand for distributed AI training

View Pricing

Frequently Asked Questions

What is Qwen3-32B FP8 good for?: Self-hosted reasoning and chat with a switch between thinking and non-thinking modes, covering over 100 languages, Alibaba says. Apache 2.0 weights, on a single high-memory GPU.
When is Qwen3-32B FP8 not a good fit?: Image, video or audio input: text only, with a short native context. It runs all 32B parameters on every token, unlike the 30B-A3B mixture of experts with 3B active, so each token takes more compute to serve.
What is the cheapest way to run Qwen3-32B FP8?: Hosted, unless you push serious volume. Alibaba Cloud charges $0.16 in / $0.64 out per 1M tokens. The cheapest rental that fits is 2x RTX 4060 Ti at $158 a month, which costs the same as about 660M tokens a month on that API.
Can I self-host Qwen3-32B FP8?: Yes. The weights are Apache 2.0 licensed. At 4-bit it needs about 24 GB of GPU memory, which starts at roughly $158 a month on the cheapest rental that fits.

More from Alibaba Cloud

Model	Context	Input /1M	Output /1M
Qwen3.6-35B-A3B Open weights	262K	$0.07	$0.42
Qwen3.5-35B-A3B Open weights	262K	$0.06	$0.46
Qwen3.5-Flash	1M	$0.10	$0.40
Qwen3-235B-A22B Instruct (2507) Open weights	262K	$0.09	$0.58
Qwen3-30B-A3B FP8 Open weights	33K	$0.05	$0.34
Qwen3.5-27B Open weights	262K	$0.09	$0.69
Qwen3-Coder-30B-A3B Open weights	262K	$0.06	$0.25
Qwen3.8-Flash	1M	$0.15	$0.47