Open weights · Apache 2.0
Qwen3-4B FP8
FP8-quantized version of Qwen3-4B, the smallest Qwen3 dense model, suited to edge deployment with a hybrid thinking mode.
Key Specifications
- Context window
- 33K tokens
- Parameters
- 4B
- 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
| Provider | Input /1M tokens | Output /1M tokens | Cost at 10M in + 2M out | |
|---|---|---|---|---|
|
Alibaba Cloud
Creator
|
Open weights, no hosted price | 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
Qwen3-4B FP8 needs about 9 GB of GPU memory at 4-bit with its full 32K context. The cheapest rental that fits is 1× RTX 3060 at about $43 a month.
| Precision | Memory | Cheapest fit (Full 32K context) | Cost /mo |
|---|---|---|---|
| 4-bitINT4 / FP4 |
9 GB
|
$43
|
|
| 8-bitFP8 / INT8 |
8 GB
|
$72
|
|
| 16-bitFP16 / BF16 |
15 GB
|
$86
|
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. No guarantee of runtime support, usable performance, or that a matching quantized build exists. How we estimate costs.
Frequently Asked Questions
What is Qwen3-4B FP8 good for?
Edge and on-device deployment, local chat and small tool-calling loops. Apache 2.0 weights at the light end of the Qwen3 lineup, small enough for a single consumer GPU.
When is Qwen3-4B FP8 not a good fit?
Image, video or audio input: it is text only. Heavier reasoning and long-document work, which the larger dense and MoE models in the same lineup are sized for.
What is the cheapest way to run Qwen3-4B FP8?
No provider we track hosts it; self-hosting on the cheapest rental that fits (1x RTX 3060, $0.06 per hour) costs about $43 a month.
Can I self-host Qwen3-4B FP8?
Yes. The weights are Apache 2.0 licensed. At 4-bit it needs about 9 GB of GPU memory, which starts at roughly $43 a month on the cheapest rental that fits.
More from Alibaba Cloud
| Model | Context | Input /1M | Output /1M |
|---|---|---|---|
|
Qwen3-Coder-Plus
|
1M | $1.00 | $5.00 |
|
Qwen3.5-Flash
|
1M | $0.10 | $0.40 |
|
Qwen3.5-Plus
|
1M | $0.40 | $2.40 |
|
Qwen3.6-Flash
|
1M | $0.25 | $1.50 |
|
Qwen3.6-Plus
|
1M | $0.50 | $3.00 |
|
Qwen3.7-Flash
|
1M | $0.03 | $0.13 |
|
Qwen3.7-Max
|
1M | $1.25 | $3.75 |
|
Qwen3.7-Plus
|
1M | $0.28 | $1.10 |