Per GPU-hour
$5.00/GPU-hr
GraphN / Pricing
Published usage pricing.
Token pricing for hosted models, GPU runtime for custom models, and platform meter rates. Numbers mirror the billing catalog and render fail-closed when a rate is unpublished.
LLM inference
USD per 1M tokens. Use built-ins for token billing, custom deployments for GPU runtime, and imported models through your provider account.
| Model | Input | Output |
|---|---|---|
| qwen3-80b | $0.20/1M | $0.60/1M |
| qwen3-235b | $0.22/1M | $0.88/1M |
| qwen3-coder | $0.90/1M | $0.90/1M |
| nemotron-3-super | $0.90/1M | $0.90/1M |
| gpt-oss-120b | $0.15/1M | $0.60/1M |
| qwen2.5-vl-7b-instruct | $0.05/1M | $0.05/1M |
| qwen3-vl | $0.10/1M | $0.15/1M |
| qwen3.8-27b | $0.45/1M | $3.20/1M |
| qwen3.5-122b-a10b-fp8 | $0.29/1M | $2.40/1M |
| gemma-4 | $0.39/1M | $0.97/1M |
| gemma-4-e4b | $0.15/1M | $0.60/1M |
| custom:* | Runs on our GPUs and is billed by runtime, not tokens. | |
| imported:* | Runs through your provider account and is not billed by GraphN. | |
Custom model GPU runtime
Per GPU-second
$0.00138889/GPU-second
Platform meters
KB embedding (BGE-M3)
$0.02/1M input tokens
The embedding model selected when the knowledge base is created determines this rate.
KB reranking (BGE-M3)
$0.02/1M input tokens
KB embedding (Qwen3-VL)
$0.11/1M input tokens
The embedding model selected when the knowledge base is created determines this rate.
KB reranking (Qwen3-VL)
$0.045/1M input tokens
Unstructured RAG ingestion
$4.00/1,000 pages
Markdown conversion
$0.50/1,000 pages
Object storage
$0.069/GB-month
Standing bytes are measured hourly; 730 hours equals one billing month.
Knowledge base vector storage
$1.10/GB-month
Qdrant vector and payload bytes are measured hourly.
Function invocation
$0.40/1M calls
Connector invocation
$0.40/1M calls
Function compute
$0.0000166667/GB-second
RAG job trigger
$0/call
Free allowance: $0.00/call
Pricing FAQ
What counts as a GPU-second?
One GPU-second is one second of active runtime on a single GPU for a custom model using your weights on GraphN-managed GPUs. Runtime is billed per GPU, so a deployment on two GPUs accrues two GPU-seconds for every wall-clock second.
Are built-in and imported models billed the same way?
No. Built-ins bill by token usage on GraphN. Imported models run on your external provider and are billed by that provider.
Where can I estimate monthly spend by workload?
In the app, go to your organization billing pricing section for calculator-style estimates and breakdown arithmetic.