Pay for what you use.
Nothing else.
No subscriptions, no minimum spend, no sales call. Every new account starts with $5 of free usage — enough to embed, forecast, and train your first model.
Call the models
Metered per request. The rate is the whole price — no platform fee, no tiers.
inertial-embed-alphaEmbeddings$0.25 / 1M tokensinertialai-forecastForecasting$0.01 / call + $0.15 / 1K pointsinertialai-forecasting-reasoningReasoning forecastForecast price + reasoning computeFinetune on your data
Training is priced on the data you train on — per million training units, with a $5 minimum per job. A first experiment costs about a coffee; a production dataset scales linearly. If the tuned model doesn't beat the base on the held-out eval, the fee refunds automatically.
inertialai-embedTuned embedding spaces for search and retrieval
chronicleThe multimodal foundation model — forecasting, classification, regression, text generation
inertialai-forecastThe production forecaster, trained on larger GPUs
A training unit is one text token; four time-series points count as one unit. Rates are for LoRA adapters — full finetunes update every weight and bill at 4× the rate. $5 minimum per job. Datasets and finetuned weights are stored free (up to 10 GB of retained datasets per account).
Deploy and serve
Finetuned models serve on a dedicated GPU, billed by GPU time at the hourly rates below. A call is metered wall-clock, cold-start included; always-warm replicas bill hourly while reserved. Scale-to-zero endpoints cost nothing between calls.
T4$0.97 / GPU-hourFirst experiments and light trafficL4$1.32 / GPU-hourThe default for Chronicle finetunesA10G$1.82 / GPU-hourProduction endpoints at steady loadL40S$3.22 / GPU-hourBigger batches, lower latencyA100 (40 GB)$3.47 / GPU-hourHeavy modelsA100 (80 GB)$4.13 / GPU-hourHeavy models, larger contextH100$6.52 / GPU-hourLatency-critical serving at scaleRates are for the auto region; pinning to a specific region bills at 1.25–1.5× the listed rate.
Full validation before payment
Every record of your file is checked — shape, modalities, labels — and you see exactly what a run will cost before paying a cent.
Before/after eval on every job
Each finetune ends with a held-out comparison against the frozen base model. If your model doesn't beat the base, the training fee refunds automatically.
Deploys, versioning, rollback
One call turns a finetune into a private endpoint with scoped keys, version history, and one-call rollback.
Zero idle cost
Scale-to-zero endpoints cost nothing between calls, and every endpoint has a hard spend cap — alert at 80%, suspend at 100%.
Cost calculator
Estimate monthly usage for any model — embeddings, forecasts, reasoning, or a deployed finetuned endpoint.
inertial-embed-alpha at $0.25 per 1M input tokens.
Running at volume?
For committed usage, single-tenant deployments, invoiced billing, or a guided 30-day pilot on your data, talk to us.