Hosted inference for open-weight models, billed per 1M tokens. Also rents dedicated GPUs, GPU clusters and fine-tuning, priced per GPU-hour.
Developers who want to call open-weight LLMs through an API billed per token, and teams that later need dedicated GPUs or fine-tuning from the same vendor.
Teams that need one flat monthly price: serverless is per token, dedicated endpoints and clusters per GPU-hour, so the bill moves with usage. Also not for anyone who needs Provisioned Throughput pricing from the page, because the per-PTU rate is not published.
Pros
Cons
Serverless inference is billed per 1M tokens, with separate input and output rates for each model. For example, gpt-oss-120B is $0.15 input and $0.60 output per 1M tokens in USD. Some models also list a lower cached-input rate.
Dedicated Inference on-demand is $5.49 per GPU-hour for an NVIDIA HGX H100 and $8.99 per GPU-hour for an HGX B200, in USD. H200, B300, GB200 NVL72 and GB300 NVL72 capacity, and all reserved dedicated capacity, require contacting sales.
The pricing page shows a 'Start for free' call to action but does not state a free credit amount or list plan tiers. Check the signup flow or the vendor directly for the current credit figure before relying on one.
Yes. Fine-tuning is priced per 1M tokens with a minimum charge per job. Llama 3.3 70B is $2.03 supervised and $5.08 for DPO per 1M tokens, with a $4.00 minimum, in USD. Rates rise steeply for larger models, such as $40.00 per 1M supervised tokens for GLM-5.2.
Yes. GPU Clusters are priced per GPU-hour by hardware and commitment. An H100 is $1.99 preemptible, $3.99 on-demand, and $3.19 to $3.69 on 3 to 180 day reservations, all in USD. GB200 and GB300 NVL72 are reserved-only through sales.
Facts sourced from the vendor’s own site, last verified 2026-10-30.
The Weekly Source
One email a week. New listings, repriced tools, and the ones we would skip - including this one if its pricing changes.
