LLM training cost calculator

Training compute costs the total node rate multiplied by runtime. Enter hours directly, or estimate them from your token count and your own throughput measurement.

Training budget formula

In hours mode: compute bill = hours × GPU count × rate per GPU-hour. In tokens mode: hours = tokens per epoch × epochs ÷ tokens per second ÷ 3,600. Throughput must describe the whole selected node, including the parallelism strategy you plan to use.

Model size alone cannot predict runtime. Optimizer state, activations, context length, precision and the training method change the memory needed. Compare inference, adapter training and full-training memory scenarios.

Rates updated 2026-10-02. This estimate excludes storage, data transfer, checkpoint overhead, evaluation, failed runs and idle time outside your runtime input. Add those hours explicitly and use the job cost calculator for published storage and egress charges.