Pay for what you train, or scope it with us

Start with $200 of credit and pay as you go. Expert-led PoCs and self-hosted deployments are scoped with you.

Self-service

Your engineers build models on the platform.

$200of credit to start, no card needed. Then pay per job.
  • Task description and as few as 10 examples to start
  • Models hosted with us on a dedicated endpoint
  • Non-commercial model licence
  • Community support on Slack

Expert-led PoC

A scoped project with acceptance criteria, priced per use case.

Per use case3 to 4 weeks, ending in a handover to your team
  • Our forward deployed engineers in your team
  • Harness integration, evals, data work and training
  • Readout against every acceptance criterion
  • Onboarding for your engineers

Self-hosted

Download the weights and run them in your own estate.

Platformannual fee plus a committed training budget
  • Your cloud, your VPC, on-premises or air-gapped
  • Full ownership of your models
  • Retraining from your own traces, on your hardware
  • SLAs, procurement and security review support

Building a model

Each job is priced at the tier of the model it runs. Larger models and thinking variants are in the higher tiers.

JobUnitLowMediumHigh
Priced at the teacher's tier, for example gpt-oss-120b, DeepSeek V4 Pro, Kimi K3
Teacher evaluationper 1M test set tokens$1$4$12
Relabeling tracesper 1M trace tokens, for each teacher$1$4$12
Synthetic data generationper 1M generated tokens$1$4$12
Priced at the student's tier, for example Qwen3 0.6B, Qwen3 4B, Qwen3.5 9B
SFT trainingper 1M train set tokens × epochs$0.10$0.25$0.50
RLVR trainingper 1M train set tokens × epochs$0.60$1.50$3
Deploymentper deployment$5$5$5

Uploads, downloads and reads are free, and a dry run shows a job's price before you start it. How billing works

A dedicated GPU, billed per hour

Your model on its own H100. You pay per GPU hour of uptime, not per token, so your effective rate falls as traffic grows.

Blended price per 1M tokens at sustained load

Your model, Qwen3 4B$0.04
Your model, Qwen3.5 9B$0.08
Gemini 2.5 Flash-Lite$0.18
GPT-5.4 Nano$0.46
Claude Haiku 4.5$2.00

Three input tokens per output token. Effective rates at sustained load on one H100. A dedicated GPU makes sense above roughly 30% utilisation.

Build your first model on us

$200 of credit, no card needed. Or talk to us about your use case.