What it actually costs, this month, to clear a quality bar — measured across real routed traffic and tune runs, per task type.
Research · As of July 2026
Cost dashboards tell you what you spent. o10 sits in the request path and changes what you spend — shadow first, then enforce.
Short, self-contained answers with cited stats — read the sections below for full context.
The Quality-Cost Frontier Index answers that per task type: the cheapest model that clears a stated quality bar (0.74 / 0.86 / 0.93) and its measured cost per 1K calls — with observation and org counts on every real row.
2 measured cells for July 2026.
Prompt content. The index is metrics-only with k-anonymity minimums on observations and orgs before a cell publishes.
No prompts · k-anonymity thresholds · public citation target
o10 holds your quality floor in the request path and routes each call to the cheapest model that clears it. The index is the public, aggregatable expression of that underwriting — savings are the consequence of clearing the bar cheaper, not a claim without proof.
How a cell enters the index — and what never does.
Quality is holdout-scored on customer-supplied examples (Tune) and enforced via eval-gated live routing. A model is listed as cheapest-clearing only when it empirically clears the stated quality bar.
Publishing requires k-anonymity minimums: both observation_count and org_count must clear internal thresholds before a cell appears. Prompt content is never collected for this index.
Costs are reported as USD per 1K calls for the clearing model at that task type × quality bar. Periods are calendar months unless the API specifies otherwise.
Source: https://app.o10.io/api/frontier-index. Page resolves live fetch → last cached period → ILLUSTRATIVE methodology preview. Never present sample rows as measured.
Cheapest clearing model and cost per 1K calls
— measured aggregates. Every row shows observation_count and org_count.
| Task type | Quality bar | Cheapest clearing model | $/1K calls | n (obs) | Orgs |
|---|---|---|---|---|---|
| classification | 0.86 | llama-3.1-8b | $0.420 | 128 | 7 |
| codegen | 0.93 | deepseek-v4-pro | $1.15 | 64 | 5 |
How to cite: o10 Quality-Cost Frontier Index, July 2026. Canonical: www.o10.io/research/quality-cost-frontier-index. Distribution: https://app.o10.io/api/frontier-index.
A privacy-safe monthly dataset from o10: for each task type and quality bar, the cheapest model that empirically clears the bar and its cost per 1K calls, with observation_count and org_count. It is designed to be cited — not a marketing estimate.
Metrics only: task type, quality bar, clearing model, cost per 1K calls, and k-anonymized observation/org counts. No prompt content is collected for this index. Rows publish only when observation and org thresholds clear.
Holdout-scored evals on customer-supplied examples (Tune) and eval-gated live routing receipts. A model only appears as cheapest-clearing when it empirically clears the bar at that quality threshold.
Yes for period 2026-07 (As of July 2026). Every row shows observation_count and org_count. Source: live.
Use: “o10 Quality-Cost Frontier Index, July 2026”. Canonical page: https://www.o10.io/research/quality-cost-frontier-index. Machine-readable distribution: https://app.o10.io/api/frontier-index.
Paste a week of traffic. Get the number that books the audit.
See what you're overpaying →