User-supplied performance only
GPU inference cost calculator
Estimate GPU cost per hour, monthly spend and cost per one million tokens. Choose an official cloud rate, then enter throughput measured on your own model, serving stack and workload. This tool contains no benchmark assumptions.
- Only rates marked live, recent or verified when the catalog was built
- No default tokens-per-second claim
- Every selectable rate links to official evidence
This catalog snapshot is older than 18 hours. Offer selection and calculations are paused; the official evidence links remain available while Guardian refreshes the data.
Inference scenario
Use your measured workload inputs
Throughput means output tokens per second for one GPU while inference is active. Utilization is the share of each billed hour spent producing tokens. Monthly hours are billable GPU-hours, capped at 744 for this planning view.
Select an offer and enter all three workload inputs. Empty fields are intentional: SaaS Sentinel does not invent inference performance.
JSON-driven rate selector
Choose an official GPU rate from this snapshot
These records were comparable, marked live, recent or verified, expressed in USD per GPU-hour, and linked to an official HTTPS source when the catalog was generated. To keep the calculator fast, the selector retains the lowest current row for each provider, GPU and market combination; all 859 regional rows remain in the global comparator and downloads.
58 representative rates shown
| Use rate | Provider / GPU | Region / market | USD / GPU-hour | Freshness | Evidence |
|---|---|---|---|---|---|
| VultrA16 · Ampere 16GB | atlon demand · 2 GPU/node | $0.4710 | recentObserved 2026-08-13 | Official source ↗ | |
| PaperspaceP4000 · Pascal 8GB | ams1on demand · 1 GPU/node | $0.5100 | recentObserved 2026-08-13 | Official source ↗ | |
| Akamai Cloud ComputingRTX 4000 Ada · Ada 20GB | ap-northeaston demand · 1 GPU/node | $0.5200 | recentObserved 2026-08-13 | Official source ↗ | |
| PaperspaceRTX 4000 · Turing 8GB | ams1on demand · 1 GPU/node | $0.5600 | recentObserved 2026-08-13 | Official source ↗ | |
| Microsoft AzureA100 · PCIe 80GB | eastusspot · 1 GPU/node | $0.6788 | recentObserved 2026-08-13 | Official source ↗ | |
| PaperspaceA4000 · Ampere 16GB | ams1on demand · 1 GPU/node | $0.7600 | recentObserved 2026-08-13 | Official source ↗ | |
| PaperspaceP5000 · Pascal 16GB | ams1on demand · 1 GPU/node | $0.7800 | recentObserved 2026-08-13 | Official source ↗ | |
| ScalewayL4 · 24GB | fr-par-1on demand · 1 GPU/node | $0.7875 | recentObserved 2026-08-13 | Official source ↗ | |
| PaperspaceRTX 5000 · Turing 16GB | ny2on demand · 1 GPU/node | $0.8200 | recentObserved 2026-08-13 | Official source ↗ | |
| OVHcloudTesla V100S · 32GB | us-east-va-1on demand · 1 GPU/node | $0.8800 | recentObserved 2026-08-13 | Official source ↗ | |
| Nebius AI CloudRTX PRO 6000 · Blackwell 96GB | us-central1spot · 1 GPU/node | $0.9500 | recentObserved 2026-08-13 | Official source ↗ | |
| OVHcloudL4 · 24GB | us-east-va-1on demand · 2 GPU/node | $1.0000 | recentObserved 2026-08-13 | Official source ↗ | |
| DataCrunch (now Verda)RTX 6000 Ada · Variant not specified | Global / unspecifiedon demand · 1 GPU/node | $1.0400 | recentObserved 2026-08-13 | Official source ↗ | |
| PaperspaceP6000 · Pascal 24GB | ams1on demand · 1 GPU/node | $1.1000 | recentObserved 2026-08-13 | Official source ↗ | |
| Microsoft AzureH100 · NVL 94GB | eastusspot · 2 GPU/node | $1.2899 | recentObserved 2026-08-13 | Official source ↗ | |
| CivoL40S · Variant not specified | Global / unspecifiedon demand · 1 GPU/node | $1.2900 | recentObserved 2026-08-13 | Official source ↗ | |
| PaperspaceA5000 · Ampere 24GB | ca1on demand · 1 GPU/node | $1.3800 | recentObserved 2026-08-13 | Official source ↗ | |
| ScalewayL40S · 48GB | pl-waw-2on demand · 1 GPU/node | $1.4699 | recentObserved 2026-08-13 | Official source ↗ | |
| HyperstackA100 · Variant not specified | Global / unspecifiedon demand · 1 GPU/node | $1.4800 | recentObserved 2026-08-13 | Official source ↗ | |
| Akamai Cloud ComputingRTX 6000 · Turing 24GB | ap-northeaston demand · 1 GPU/node | $1.5000 | recentObserved 2026-08-13 | Official source ↗ | |
| Crusoe CloudL40S · Variant not specified | Global / unspecifiedon demand · 1 GPU/node | $1.5000 | recentObserved 2026-08-13 | Official source ↗ | |
| DigitalOceanL40S · Variant not specified | Global / unspecifiedon demand · 1 GPU/node | $1.5700 | recentObserved 2026-08-13 | Official source ↗ | |
| DigitalOceanRTX 6000 Ada · Variant not specified | Global / unspecifiedon demand · 1 GPU/node | $1.5700 | recentObserved 2026-08-13 | Official source ↗ | |
| VultrL40S · Variant not specified | Global / unspecifiedon demand · 1 GPU/node | $1.6700 | recentObserved 2026-08-13 | Official source ↗ | |
| CivoA100 · Variant not specified | Global / unspecifiedon demand · 1 GPU/node | $1.7900 | recentObserved 2026-08-13 | Official source ↗ | |
| Nebius AI CloudRTX PRO 6000 · Blackwell 96GB | us-central1on demand · 1 GPU/node | $1.8000 | recentObserved 2026-08-13 | Official source ↗ | |
| OVHcloudL40S · 48GB | us-east-va-1on demand · 2 GPU/node | $1.8000 | recentObserved 2026-08-13 | Official source ↗ | |
| PaperspaceA6000 · Ampere 48GB | ca1on demand · 1 GPU/node | $1.8900 | recentObserved 2026-08-13 | Official source ↗ | |
| Nebius AI CloudH100 · NVLink 80GB | eu-north1spot · 1 GPU/node | $2.1500 | recentObserved 2026-08-13 | Official source ↗ | |
| Crusoe CloudA100 · Variant not specified | Global / unspecifiedon demand · 1 GPU/node | $2.3000 | recentObserved 2026-08-13 | Official source ↗ | |
| PaperspaceV100 · 32GB | ny2on demand · 1 GPU/node | $2.3000 | recentObserved 2026-08-13 | Official source ↗ | |
| LambdaA100 · Variant not specified | Global / unspecifiedon demand · 1 GPU/node | $2.3900 | recentObserved 2026-08-13 | Official source ↗ | |
| Nebius AI CloudH200 · NVLink 141GB | eu-north1spot · 1 GPU/node | $2.4500 | recentObserved 2026-08-13 | Official source ↗ | |
| HyperstackH100 · Variant not specified | Global / unspecifiedon demand · 1 GPU/node | $2.8500 | recentObserved 2026-08-13 | Official source ↗ | |
| ScalewayH100-PCIe · 80GB | fr-par-2on demand · 1 GPU/node | $2.8665 | recentObserved 2026-08-13 | Official source ↗ | |
| CivoH100 · Variant not specified | Global / unspecifiedon demand · 1 GPU/node | $2.9900 | recentObserved 2026-08-13 | Official source ↗ | |
| PaperspaceA100 · 40GB | ny2on demand · 1 GPU/node | $3.0900 | recentObserved 2026-08-13 | Official source ↗ | |
| ScalewayH100-SXM · 80GB | fr-par-2on demand · 2 GPU/node | $3.3099 | recentObserved 2026-08-13 | Official source ↗ | |
| Microsoft AzureA100 · SXM 40GB | eastuson demand · 8 GPU/node | $3.3996 | recentObserved 2026-08-13 | Official source ↗ | |
| CivoH200 · Variant not specified | Global / unspecifiedon demand · 1 GPU/node | $3.4900 | recentObserved 2026-08-13 | Official source ↗ | |
| LambdaH100 · Variant not specified | Global / unspecifiedon demand · 1 GPU/node | $3.7900 | recentObserved 2026-08-13 | Official source ↗ | |
| Nebius AI CloudH100 · NVLink 80GB | eu-north1on demand · 1 GPU/node | $3.8500 | recentObserved 2026-08-13 | Official source ↗ | |
| Crusoe CloudH100 · Variant not specified | Global / unspecifiedon demand · 1 GPU/node | $3.9000 | recentObserved 2026-08-13 | Official source ↗ | |
| Nebius AI CloudB200 · NVLink 180GB | me-west1spot · 1 GPU/node | $3.9500 | recentObserved 2026-08-13 | Official source ↗ | |
| HyperstackH200 · Variant not specified | Global / unspecifiedon demand · 1 GPU/node | $3.9900 | recentObserved 2026-08-13 | Official source ↗ | |
| DataCrunch (now Verda)H200 · Variant not specified | Global / unspecifiedon demand · 1 GPU/node | $4.0000 | recentObserved 2026-08-13 | Official source ↗ | |
| Crusoe CloudH200 · Variant not specified | Global / unspecifiedon demand · 1 GPU/node | $4.2900 | recentObserved 2026-08-13 | Official source ↗ | |
| Nebius AI CloudB300 · NVLink 288GB | eu-west2spot · 1 GPU/node | $4.3000 | recentObserved 2026-08-13 | Official source ↗ | |
| DigitalOceanH100 · Variant not specified | Global / unspecifiedon demand · 1 GPU/node | $4.4100 | recentObserved 2026-08-13 | Official source ↗ | |
| DigitalOceanH200 · Variant not specified | Global / unspecifiedon demand · 1 GPU/node | $4.4700 | recentObserved 2026-08-13 | Official source ↗ | |
| Nebius AI CloudH200 · NVLink 141GB | eu-north1on demand · 1 GPU/node | $4.5000 | recentObserved 2026-08-13 | Official source ↗ | |
| PaperspaceH100 · SXM 80GB | ny2on demand · 1 GPU/node | $5.9500 | recentObserved 2026-08-13 | Official source ↗ | |
| VerdaB200 · Variant not specified | Global / unspecifiedon demand · 1 GPU/node | $6.1100 | recentObserved 2026-08-13 | Official source ↗ | |
| Microsoft AzureH100 · NVL 94GB | eastuson demand · 2 GPU/node | $6.9800 | recentObserved 2026-08-13 | Official source ↗ | |
| Nebius AI CloudB200 · NVLink 180GB | me-west1on demand · 1 GPU/node | $7.1500 | recentObserved 2026-08-13 | Official source ↗ | |
| ScalewayB300-SXM · 288GB | fr-par-2on demand · 8 GPU/node | $7.5000 | recentObserved 2026-08-13 | Official source ↗ | |
| Nebius AI CloudB300 · NVLink 288GB | eu-west2on demand · 1 GPU/node | $7.8500 | recentObserved 2026-08-13 | Official source ↗ | |
| CoreWeaveB200 · Variant not specified | Global / unspecifiedon demand · 1 GPU/node | $8.6000 | recentObserved 2026-08-13 | Official source ↗ |
No eligible rate matches all three filters. Remove one filter to continue.
Transparent calculation
What the result includes—and what it does not
Monthly spend
official GPU-hour rate × billable hoursThe selected catalog rate is already normalized to one GPU-hour. A provider may bill a larger node or impose other minimums; check its source.
Effective output
tokens/sec × utilization × 3,600 × hoursThroughput and utilization come entirely from you. They should reflect output tokens, not prompts, and the same workload definition across scenarios.
Cost per million
monthly GPU cost ÷ monthly tokens × 1,000,000Hours cancel mathematically in this unit cost, but remain visible because they determine monthly spend and token volume.
Excluded: CPUs bundled with a node, storage, network egress, orchestration, model hosting software, support, taxes, idle capacity outside the utilization estimate, failed requests, prompt-token cost and provider availability. This is a planning calculation, not a performance benchmark or purchasing claim.