DATA · Last updated 2026-07-24 · Shen Pandi
Pricing index
The pricing index lists representative frontier and open-weight hosted model prices in USD per million tokens, stamped with an as-of date, so private equity operating partners can compare unit economics before they commit portfolio inference spend.
- Prices as of 2026-07-01 — cite the stamp whenever you reuse a figure.
- These are reference rates for planning, not your negotiated invoice.
- Pair with quality floors and the inference calculator before routing traffic.
As-of date
All unit prices on this page are as of 2026-07-01. Vendor list prices change; treat any copy without a date as unsafe for board use. For live scenario math, open the inference cost calculator.
Frontier models
Open-weight (hosted) models
Hosted open-weight endpoints are shown because most mid-market portcos consume them via API rather than self-hosting GPUs. Self-host total cost of ownership needs a separate engineering model. Decision framework: open-weight vs frontier.
Methodology
We maintain a small set of named model classes that map to commonly procured tiers, not every SKU in the market. Input and output prices are stored separately because workloads differ in output intensity — chat assistants are not document extraction. The calculator applies an input-share assumption per workload to produce a blended rate.
Comparisons assume eval-gated routing: a cheaper model should clear a per-workload quality floor before it wins traffic. Prices are not quotes, do not include committed-use discounts, and exclude ancillary costs (guardrails, logging, embeddings, retries). Editorial independence and correction policy: About. Deeper FinOps practice: inference costs.
Frequently asked questions
What is the pricing index?
The pricing index is a reference table of representative frontier and open-weight hosted model unit prices (USD per 1M input and output tokens) with an explicit as-of date for portfolio planning.
Are these negotiated rates?
No. Figures are illustrative list-style reference rates for comparison and calculators. Your contracted prices may differ. Always label board packs with as-of dates.
How should PE teams use these numbers?
Combine unit rates with workload token estimates and routing assumptions. Use the inference calculator for scenarios, and require quality floors before shifting traffic to cheaper models.
How often is the index updated?
When vendor list prices move materially or at least when we republish the as-of date. Check the as-of stamp on this page before citing.