// tools/gpu-cost-calculator
GPU Cost Calculator
Compare every managed API and self-hosted GPU for AI image generation. See exactly when self-hosting beats the APIs — and by how much. 38 GPU options · 15 managed APIs · prices verified June 2026
images
Don't want to manage GPU infrastructure? Set the DevOps overhead to $0 above — that's what you'd pay on Runflow: compute-only pricing on a managed ComfyUI platform. No engineer needed for scaling, monitoring, or cold starts. Full ComfyUI platform comparison →
📡 Managed API wins at this volume
With $3,000/mo in DevOps costs, self-hosting (Salad RTX 3090) breaks even at 198K images/month. At 10K/mo you're below that — managed API is cheaper total.
Cheapest API
$154
Together AI / mo
Cheapest GPU + ops
$3,003
Salad RTX 3090
📡 Managed APIs — zero infrastructure
| # | Provider | $/image | 10K/mo |
|---|---|---|---|
| 1 | Together AI ✓ 3 months free for new developers | $0.015 | $154 |
| 2 | fal.ai | $0.025 | $250 |
| 3 | Replicate | $0.025 | $250 |
Prices at 1024×1024 · verified June 2026 · full breakdown →
⚡ Self-Hosted GPUs — compute only
| # | Provider · GPU | $/hr | ~/image | compute |
|---|---|---|---|---|
| 1 | Salad RTX 3090 · 24GB | $0.090 | ~$0.00028 | $2.81 |
| 2 | Salad RTX 4090 · 24GB | $0.200 | ~$0.00036 | $3.64 |
| 3 | RunPod RTX 4090 · 24GB | $0.340 | ~$0.00062 | $6.18 |
| 4 | Salad H100 80GB · 80GB | $0.990 | ~$0.00062 | $6.19 |
| 5 | TensorDock RTX 3090 · 24GB | $0.200 | ~$0.00063 | $6.25 |
| 6 | TensorDock RTX 4090 · 24GB | $0.360 | ~$0.00065 | $6.55 |
| 7 | Vast.ai RTX 3090 · 24GB | $0.240 | ~$0.00075 | $7.50 |
| 8 | Vast.ai L40S · 48GB | $0.470 | ~$0.00078 | $7.83 |
| 9 | TensorDock A100 80GB · 80GB | $0.750 | ~$0.00079 | $7.89 |
| 10 | Vast.ai RTX 5090 · 32GB | $0.870 | ~$0.00079 | $7.91 |
| 11 | Thunder Compute A100 80GB · 80GB | $0.780 | ~$0.00082 | $8.21 |
| 12 | Vast.ai H100 80GB · 80GB | $1.38 | ~$0.00086 | $8.63 |
| 13 | Thunder Compute H100 80GB · 80GB | $1.38 | ~$0.00086 | $8.63 |
| 14 | Vast.ai RTX 4080 · 16GB | $0.330 | ~$0.00087 | $8.68 |
| 15 | RunPod RTX 5090 · 32GB | $0.990 | ~$0.00090 | $9.00 |
| 16 | Vast.ai A40 · 48GB | $0.290 | ~$0.00091 | $9.06 |
| 17 | Vast.ai A100 80GB · 80GB | $0.901 | ~$0.00095 | $9.48 |
| 18 | Vast.ai RTX 4090 · 24GB | $0.670 | ~$0.0012 | $12 |
| 19 | Vast.ai RTX A6000 · 48GB | $0.390 | ~$0.0012 | $12 |
| 20 | RunPod A100 80GB · 80GB | $1.19 | ~$0.0013 | $13 |
| 21 | Vast.ai H100 NVL · 94GB | $2.52 | ~$0.0014 | $14 |
| 22 | TensorDock RTX A6000 · 48GB | $0.450 | ~$0.0014 | $14 |
| 23 | TensorDock H100 80GB · 80GB | $2.25 | ~$0.0014 | $14 |
| 24 | RunPod L40S · 48GB | $0.860 | ~$0.0014 | $14 |
| 25 | RunPod RTX 3090 · 24GB | $0.460 | ~$0.0014 | $14 |
| 26 | RunPod RTX A6000 · 48GB | $0.490 | ~$0.0015 | $15 |
| 27 | RunPod H100 80GB · 80GB | $2.49 | ~$0.0016 | $16 |
| 28 | Vast.ai H200 SXM · 141GB | $3.88 | ~$0.0017 | $17 |
| 29 | Crusoe A100 80GB · 80GB | $1.80 | ~$0.0019 | $19 |
| 30 | RunPod H200 SXM · 141GB | $4.39 | ~$0.0019 | $19 |
| 31 | Modal A100 80GB · 80GB | $2.10 | ~$0.0022 | $22 |
| 32 | Crusoe H100 80GB · 80GB | $3.90 | ~$0.0024 | $24 |
| 33 | Modal H100 80GB · 80GB | $3.95 | ~$0.0025 | $25 |
| 34 | Lambda H100 80GB · 80GB | $3.99 | ~$0.0025 | $25 |
| 35 | CoreWeave A100 80GB · 80GB | $2.70 | ~$0.0028 | $28 |
| 36 | Lambda A100 80GB · 80GB | $2.79 | ~$0.0029 | $29 |
| 37 | Lambda RTX A6000 · 48GB | $1.09 | ~$0.0034 | $34 |
| 38 | CoreWeave H100 80GB · 80GB | $6.16 | ~$0.0039 | $39 |
Throughput estimated · batch mode (run only while generating) · no DevOps overhead included above
Methodology: GPU costs assume batch mode — you rent for exactly as long as it takes to generate your images. Throughput is estimated from single-GPU benchmarks at 1024×1024. DevOps overhead is the monthly cost of engineering time to manage GPU infrastructure (monitoring, autoscaling, driver updates). API prices are flat per-image at 1024×1024.
Prices sourced directly from provider pricing pages · More GPU cost breakdowns → · Updated June 2026