Modal Pricing 2026

Modal charges per second of active compute. GPU pricing ranges from $1.10/hr for A10G/T4 to $4.29/hr for H100 SXM. CPU is $0.000306/vCPU-second. All new accounts get $30/month in free compute credits.

Yes — Modal's Starter plan includes $30/month in compute credits at no cost. These credits apply to any GPU or CPU workload, making it free for light personal use and experimentation.

Modal bills per millisecond of actual execution. There are no idle charges — you only pay when your function is actively running. This makes Modal very cost-effective for sporadic workloads compared to always-on servers.

Modal supports T4, A10G, L40S, A100 (40GB PCIe and 80GB SXM), H100 (PCIe and SXM), and H200. GPU availability varies; H100 and H200 require selection via the SDK.

For sporadic/bursty workloads, Modal is cheaper due to zero idle charges. For sustained heavy usage, RunPod's Secure Cloud or Community Cloud (spot at ~50% off) is cheaper per hour. Modal A10G costs $1.10/hr vs RunPod A10G at ~$0.69/hr on-demand — but Modal's billing precision eliminates wasted compute.

The plan fee ($0 for Starter, $250/month for Team) is a platform access charge only. GPU compute is billed separately per-hour on top of the plan. Baseline GPU costs start around $2/hr per GPU, H100s run ~$4.5/hr, and serving large models can reach ~$72/hr. Actual monthly spend depends entirely on compute utilization.

Yes. Modal's serverless architecture spins containers on-demand rather than keeping GPUs always-on. This produces cold start latency when functions haven't been called recently. Users also need to configure their own inference stack (e.g., vLLM) rather than using a managed inference endpoint.

Based on community comparisons, Modal's H100 runs ~$4.5/hr versus RunPod's H100 at ~$2.5/hr. RunPod also offers lower-tier GPUs like RTX A5000 and RTX 3090 at ~$0.22/hr. Modal's higher pricing reflects its managed serverless infrastructure and simpler deployment model. Check modal.com/pricing directly for current GPU-specific rates.

This is a common question. Modal's on-demand model is designed around paying for execution time rather than idle GPU time, which is its core value proposition versus always-on GPU providers. However, confirm the specific billing model for your workload type at modal.com/pricing, as container warm-up and minimum billing increments may apply.

No. The Starter plan has a $0 platform fee, but GPU compute is billed separately at pay-as-you-go rates on top of the plan cost. Every hour of GPU usage is charged regardless of which plan you are on. The free plan covers access to the platform, not the compute itself.

Community reports cite baseline GPU rates starting around $2/hour per GPU for standard instances. Serving large models such as Kimi K2 can cost approximately $72/hour. Costs scale with GPU type, model size, and sustained usage hours — multiple users have noted that costs escalate rapidly for continuous production workloads.

The Starter plan has no platform fee and is suited for solo developers. The Team plan costs $250/month and adds multi-user collaboration features for team-based workflows. Both plans charge GPU compute separately at pay-as-you-go rates. Enterprise is custom-priced with dedicated SLAs and support.

Users have reported difficulty deleting accounts when an outstanding balance exists. Support must be contacted to resolve the balance before account deletion is permitted, which some users describe as creating a billing loop where charges continue while they cannot exit.

Modal is positioned as a low-cost but DIY option — you get on-demand GPU access but must configure your own inference stack (e.g., vLLM) and manage cold starts yourself. It suits teams with infrastructure expertise who want flexibility and pay-as-you-go billing. Teams wanting turnkey managed inference with guaranteed latency SLAs may find fully managed providers a better fit.