Runpod Serverless Flex pricing: why VRAM alone is the wrong comparison
Runpod's Flex tiers range from $0.58/hour 16GB classes to high-end B300 capacity. The useful decision is model fit, hardware class, and measured cost per job—not VRAM alone.
Runpod Serverless Flex lists NVIDIA A100 80GB at USD 2.72 per GPU-hour, billed per second for GPU worker time including startup and idle periods. Failed or timed-out jobs can incur charges. Separately, queue-based Batch Inference became generally available on October 28, 2026, with GPU-dependent discounts up to 30% versus on-demand; an A100-specific batch rate is not confirmed.
Secondary price terms only; primary pricing stays in the deal summary.
Use-case fit and limitations from the existing resource record.
Serverless premium may not be economical for sustained high-utilization workloads compared with dedicated GPU instances. Failed/timed-out/retried jobs and worker startup/idle time may still incur GPU-second charges.
Available
Decision visuals remain traceable to the existing source and verification record.
Independent ranking · native prices · evidence-linked · commercial relationships isolated from scoring and publication.