Small
Up to 13B params
~$1.50/hr
Llama 3.1 8B, Qwen 3.5 9B
Pricing
Transparent pricing.
No hidden fees.
Pick a model, deploy in minutes, pay by the hour. 1-hour minimum per deployment. Unused whole hours are refunded to your account balance.
Medium
20B – 40B params
~$5.00/hr
Qwen 3.5 27B, Qwen 3.5 35B, GPT OSS 20B
Large
70B – 130B params
~$11.50/hr
Llama 3.3 70B, Qwen 3.5 122B, Qwen3 Coder Next
XL / MoE
200B+ params
~$17.00/hr
Llama 4 Scout, MiniMax M2.5
Exact pricing depends on model. See the full catalog below.
Full model catalog
Every row is a deployable configuration.
Llama
| Model name | Parameters | Max context length | Price / hr |
|---|---|---|---|
| Llama 3.1 8B Instruct | 8B | 128k | $1.50 |
| Llama 3.3 70B Instruct | 70B | 128k | $12.00 |
| Llama 4 Scout 17B-16E Instruct | 109B (17 active) | 10M | $17.00 |
Qwen
| Model name | Parameters | Max context length | Price / hr |
|---|---|---|---|
| Qwen 3.5 9B | 9B | 256k | $1.50 |
| Qwen 3.5 27B | 27B | 256k | $6.00 |
| Qwen 3.5 35B | 35B | 256k | $6.00 |
| Qwen 3.5 122B (FP8) | 125B | 256k | $13.00 |
| Qwen3 Coder Next | 80B | 256k | $12.00 |
| Qwen3 Coder Next (FP8) | 80B | 256k | $6.50 |
GPT OSS
| Model name | Parameters | Max context length | Price / hr |
|---|---|---|---|
| GPT OSS 20B | 22B | 128k | $3.50 |
| GPT OSS 120B | 120B | 128k | $6.50 |
MiniMax
| Model name | Parameters | Max context length | Price / hr |
|---|---|---|---|
| MiniMax M2.5 | 229B | 200k | $17.00 |
Bring Your Own Model
Have a fine-tuned or custom model? Contact us and we'll help you deploy it on Inferfly. Pricing is custom based on your GPU tier. Volume and term discounts available.
| GPU tier | VRAM | Max context length | Price / hr |
|---|---|---|---|
| L40S | 48 GB | Custom | Custom |
| A100 SXM | 80 GB | Custom | Custom |
| H100 SXM | 80 GB | Custom | Custom |
| H200 SXM | 141 GB | Custom | Custom |
| B200 | 180 GB | Custom | Custom |
Disclaimer: The above prices are for reference only. Exact pricing depends on model, replica sets, number of concurrent users, and GPU type and no. of GPUs.
How does billing work?
You pay by the hour, prepaid, with a 1-hour minimum per deployment. When you stop a deployment, any unused whole hours are refunded to your account balance.
Can I withdraw my account balance?
Account balance can be used toward future deployments but cannot be withdrawn as cash.
What happens if my deployment crashes?
If a deployment fails due to an infrastructure issue on our end, the affected hours are refunded to your account balance automatically.
Do prices change based on how many requests I send?
No. You pay for compute time, not per request or per token. Once your endpoint is live, you can send as many requests as the GPU can handle.
Can I run multiple models on the same GPU?
No. Each deployment is a dedicated GPU (or set of GPUs) running a single model. This ensures predictable latency and no noisy-neighbor issues.
What does "pre-validated GPU configuration" mean?
We test every model-GPU pairing before adding it to the catalog. You won't hit out-of-memory errors or need to guess quantization settings. The config that ships is the one that works.
Can I deploy a fine-tuned or custom model?
Yes. Contact us with your model details (parameter count, base architecture, quantization format) and we will set up a dedicated deployment with custom pricing.
Is there a free trial?
Not currently. We offer a 1-hour minimum so you can validate your setup at low cost before committing to longer runs.
How fast does my endpoint go live?
Most models are live within 5 minutes of deployment. Larger models that require multi-GPU setups may take slightly longer.
Do you offer volume discounts?
Yes. If you are running sustained workloads or multiple deployments, contact us for custom pricing.
Enterprise
Need something bigger?
Looking to run Llama 4 Maverick, Qwen 3.5 397B, Kimi K2.5, or other large-scale models? We provision these on demand. Tell us what you need and we'll get you a dedicated deployment with custom pricing.
Talk to us