GPU Cloud and LLM API Pricing.
Flexible plans. Clear value.
No hidden fees, no surprises. Just transparent pricing designed to scale with your needs. Whether you’re just starting or expanding fast, there’s a solution for you.
LLM API pricing (per million tokens)
| Model | Input | Output | Cached |
|---|---|---|---|
| Kimi-K3 | $5/M tokens | $15/M tokens | $0.3/M tokens |
| Kimi-K2.6 | $1.2/M tokens | $4.5/M tokens | $0.2/M tokens |
| Kimi-K2.5 | $0.6/M tokens | $3.0/M tokens | $0.2/M tokens |
| MiniMax-M3 | $0.3/M tokens | $1.2/M tokens | $0.06/M tokens |
| MiniMax-M2.7 | $0.4/M tokens | $1.5/M tokens | $0.15/M tokens |
| MiniMax-M2.5 | $0.3/M tokens | $1.2/M tokens | $0.2/M tokens |
| Nemotron-3-Ultra-550B-A55B | $0.6/M tokens | $3.5/M tokens | $0.2/M tokens |
| Qwen3-235B-A22B-Instruct | $0.25/M tokens | $1.0/M tokens | $0.2/M tokens |
| Qwen3.6-35B-A3B | $0.25/M tokens | $1.5/M tokens | $0.2/M tokens |
| GLM-5.3 | $1.4/M tokens | $4.4/M tokens | $0.3/M tokens |
| GLM-5.2 | $1.3/M tokens | $4.2/M tokens | $0.25/M tokens |
| GLM-5.1 | $1.4/M tokens | $4.4/M tokens | $0.3/M tokens |
| GLM-5 | $1.0/M tokens | $3.0/M tokens | $0.2/M tokens |
| GLM-4.7 | $0.7/M tokens | $2.8/M tokens | $0.2/M tokens |
| Deepseek-V4-Pro | $1.7/M tokens | $3.4/M tokens | $0.15/M tokens |
| Deepseek-V4.1-Flash | $0.3/M tokens | $1.2/M tokens | $0.075/M tokens |
| Deepseek-V4-Flash | $0.2/M tokens | $0.4/M tokens | $0.05/M tokens |
| Deepseek-R1 | $1.5/M tokens | $4.5/M tokens | $0.3/M tokens |
| Deepseek-V3.2 | $0.5/M tokens | $1.5/M tokens | $0.15/M tokens |
| Gemma-4-31B-IT | $0.35/M tokens | $0.8/M tokens | $0.2/M tokens |
| Gemma-4-26B-A4B-IT | $0.2/M tokens | $0.5/M tokens | $0.1/M tokens |
| GPT-OSS 120B | $0.15/M tokens | $0.6/M tokens | $0.1/M tokens |
| GPT-OSS 20B | $0.07/M tokens | $0.25/M tokens | $0.05/M tokens |
| Llama 3.3 70B Instruct | $0.7/M tokens | $0.7/M tokens | $0.2/M tokens |
| Llama 3.1 70B Instruct | $0.8/M tokens | $0.8/M tokens | $0.16/M tokens |
| Llama 3.1 8B Instruct | $0.15/M tokens | $0.15/M tokens | $0.03/M tokens |
| Mistral Medium 3.5 | $1.5/M tokens | $7.5/M tokens | $0.3/M tokens |
| Mistral Small 4 | $0.15/M tokens | $0.6/M tokens | $0.015/M tokens |
| Ministral 14B | $0.2/M tokens | $0.2/M tokens |
Image generation pricing
| Model | Price per image | Images per $1 |
|---|---|---|
| Flux Schnell | $0.003 | 333 |
| Flux Dev | $0.025 | 40 |
| Stable Diffusion XL | $0.005 | 200 |
Cloud GPU pricing (per hour, billed per second)
| GPU type | Price per GPU | CPU | RAM | VRAM |
|---|---|---|---|---|
| NVIDIA B300 | $8.99/hr | 30 | 275 GB | 288 GB |
| NVIDIA B200 | $7.99/hr | 30 | 184 GB | 180 GB |
| NVIDIA H200 | $5.99/hr | 44 | 182 GB | 141 GB |
| NVIDIA H100 | $3.99/hr | 32 | 185 GB | 80 GB |
| RTX PRO 6000 | $2.49/hr | 30 | 90 GB | 96 GB |
| NVIDIA A100 | $2.29/hr | 22 | 120 GB | 80 GB |
| NVIDIA L40S | $1.79/hr | 20 | 60 GB | 48 GB |
| RTX 6000 ADA | $1.49/hr | 10 | 60 GB | 48 GB |
| RTX A6000 | $0.99/hr | 10 | 60 GB | 48 GB |
Serverless GPU pricing (billed per second)
| GPU type | Price per GPU | |
|---|---|---|
| NVIDIA B300 | $9.99/hr | 288 GB |
| NVIDIA B200 | $8.99/hr | 180 GB |
| NVIDIA H200 | $6.99/hr | 141 GB |
| NVIDIA H200 | $4.99/hr | 80 GB |
| RTX PRO 6000 | $2.99/hr | 96 GB |
| NVIDIA L40S | $1.99/hr | 48 GB |
CPU instance pricing
| CPU type | vCPU | RAM | Price per hour |
|---|---|---|---|
| AMD EPYC | 4-360 | 16-1440GB | from $0.16 |
Storage pricing
| Storage type | Bandwidth | IOPS | Price per GB |
|---|---|---|---|
| NVMe | 2000 MB/s | 100k | $0.2/month |
Prices in USD.
Reserved capacity
Reserved GPUs and GPU clusters
Need guaranteed capacity for training or high-volume inference? Reserve GPUs or a dedicated multi-node cluster and get a rate based on GPU type, cluster size and commitment length.
✓ Commitment from 3 to 36 months
✓ B300, GB300 and RTX PRO 6000 with InfiniBand networking
✓ Managed Kubernetes or bare metal deployment, single tenant
Pricing FAQ
How is GPU usage billed?
Cloud GPU instances and Serverless GPU endpoints are billed per second, at the hourly rate of the GPUs you use: Cloud GPU instances from start to stop, Serverless GPU endpoints only while workers are running, with scale-to-zero. The Models API is billed per million input, output and cached tokens, or per image, and only for successful requests.
How much does an NVIDIA H100 cost per hour?
An on-demand H100 costs $3.99 per GPU per hour, billed per second, with 32 vCPUs and 185 GB of RAM included. Instances scale from 1 to 8 GPUs.
How much does an NVIDIA B300 cost per hour?
An on-demand B300 with 288 GB of HBM3e costs $8.99 per GPU per hour, with 30 vCPUs and 275 GB of RAM included. On Serverless GPU, a B300 endpoint costs $9.99 per hour, billed per second. For multi-node B300 clusters, request a quote.
Is there a free tier?
Yes. The console playground lets you test models for free: 10 requests over 10 days. After that, you only pay for what you use, with no subscription.
Do you offer volume discounts?
Yes. Rates decrease with commitment: reserved GPUs and GPU clusters from 3 to 36 months, and custom enterprise pricing for high volumes. Contact our sales team to discuss your needs.
Are there any setup fees or minimum commitments?
No setup fees and no minimum commitment on pay-as-you-go. Reserved GPUs and clusters come with a commitment in exchange for lower rates.
How do you protect my data and models?
Our own NVIDIA B300 and RTX PRO 6000 servers run in data centers in Paris and Lyon, France. Your prompts and outputs are never used to train models, your data is never shared with other customers, and each customer environment is isolated.