CMC

TX · Utils

Accelerator Cost Comparison

Hourly $/GPU across Colab, Vast.ai, Lambda, Modal, and RunPod for accelerators that show up in at least three of those catalogs — plus Colab TPUs. Full provider catalogs below.

accel.cloud($/hr)
Accelerator VRAM Peak BF16 Colab Vast Lambda Modal RunPod
Loading…
TPUs first, then GPUs by peak dense BF16 FLOPs (fastest → slowest). Inclusion rule: GPU family appears in ≥3 of Colab / Vast / Lambda / Modal / RunPod (TPUs always kept). All cost columns are $/hr. Colab = CU/hr × ~$0.10/CU; Vast = lowest verified on-demand 1×; Lambda = Instance 1×; Modal = GPU Task ($/sec × 3600; CPU/memory extra); RunPod = Community Cloud Pod. Hover a cost cell for the provider SKU name. Dates: Colab Jul 27, 2026 · Vast Jul 28, 2026 · RunPod & Modal Aug 3, 2026 · Lambda Aug 12, 2026.

Lambda Cloud Instances

On-demand Instance catalog from lambda.ai/pricing — 1× GPU rates (not 1-Click Clusters). Sorted cheapest first.

lambda.instances($50)
GPU VRAM $/hr $50 time
Loading…
Rates from Aug 12, 2026 (1× GPU view). Plus applicable sales tax/VAT/GST. A100 SXM and PCIe are both 40 GB on this snapshot. Billed per minute while running.

Modal GPU Tasks

Serverless GPU catalog from modal.com/pricing — $/hr converted from published $/sec for comparison. Sorted cheapest first. CPU and memory billed separately.

modal.gpus($50)
GPU VRAM $/hr $50 time
Loading…
Rates from Aug 3, 2026. GPU-only ($/sec × 3600); hover $/hr for the source $/sec. Starter includes $30/mo credits; Team $100/mo. Sandbox/Notebooks pricing differs.

RunPod Community Cloud Pods

On-demand Pod catalog from runpod.io/pricing — Community and Secure Cloud. Excludes Serverless, Clusters, and Reserved. Sorted by Community $/hr. $50 time uses Community.

runpod.pods($50)
GPU VRAM Community Secure $50 time
Loading…
Rates from Aug 3, 2026. Community is usually cheaper; L4 is an exception here (Secure $0.39 vs Community $0.44). Disk and network storage are extra.

References