# Pricing transcribed from the provider's own page — see provenance. Numbers are quoted, not derived.
generated: '2026-10-07'
method: searched
source: https://vast.ai/pricing
sources:
- https://vast.ai/pricing
provenance:
  tool: all/0-working/economics/harvest_pricing.py
  run: 20261007T104522
  grounding: every price_text was found verbatim in the fetched page text
audience: developer-api
pricing_basis: usage-other
delivery:
  delivery: managed
  method: page
  read: '2026-10-07'
  managed_tier: On-Demand
published: false
note: No numeric price is published on the pricing page; Page describes live market-based GPU pricing
  tiers but the visible text contains no numeric rates; live price cards and calculator did not render.
plan_count: 3
plans:
- name: On-Demand
  price_text: Per-second billing
  period: per-use
  unit: GPU-second
  included_quota: Guaranteed uptime. Best for production.
- name: Interruptible
  price_text: 50%+ cheaper.
  period: per-use
  unit: GPU-second
  included_quota: Best for batch training. Preemptible — may be reclaimed
- name: Reserved
  price_text: Contact Sales
  period: month
  included_quota: 1, 3, or 6 month terms
x-canonical:
  generated: '2026-10-08'
  vocabulary: all/0-working/economics/plans-canonical.yml
  models:
  - metered
  units:
  - compute
  plans:
  - name: On-Demand
    access: other
    model: metered
    model_how: derived
    units:
    - compute
    periods:
    - other
  - name: Interruptible
    access: other
    model: metered
    model_how: derived
    units:
    - compute
    periods:
    - other
  - name: Reserved
    access: other
    model: unstated
    model_how: derived
    units: []
    periods:
    - month