Modal Pricing in May 2026

Modal's pricing follows usage-based per-second model: $0.000725/second on Nvidia A100 (40GB), $0.001725/second on H100 (80GB), $0.000113/second on L40S, with CPU-only at $0.000040/second. Real-world Modal bills for AI products land at $50-500/month for early-stage; mid-stage products hit $5K-30K/month at scale.

The Free Credits offer provides $30 in monthly credits for new accounts. Eligibility is broad. The strategic decision: Modal's value is the serverless cloud for AI/ML — run any code on GPU or CPU without managing infrastructure, with Python-native SDK and pricing per second. For ML engineering teams who want to run arbitrary Python code with GPU access without managing Docker, Kubernetes, or VMs, Modal is uniquely positioned. The free credits are the right time to validate whether Modal's specific approach matches your ML workload. For broader serverless GPU and ML platforms, Replicate targets model deployment with broader catalog and simpler API. Together AI, Fireworks AI, and Groq target LLM inference specifically. RunPod targets cheaper GPUs with more DIY operational overhead. Lepton AI targets enterprise-scale custom models. Baseten targets custom-model deployment with strong production features. Anyscale targets Ray-based distributed ML. For specifically LLM inference, Together/Fireworks/Groq typically win on cost and latency for hosted open-weight models. For specifically Stable Diffusion and image-generation, Replicate, Fal.ai, and Stability AI compete. For specifically custom-model training and fine-tuning, Modal, Lambda Labs, and major-cloud GPUs (AWS, GCP) compete. For specifically batch inference at scale, Modal, Anyscale, and major-cloud batch platforms compete. Modal's specific value proposition is the Python-native serverless GPU.