Free prompt optimization. Send a prompt and 5–20 real examples — o10 finds a cheaper model and a tuned prompt that still clear your quality bar, and hands you a before/after receipt.
Live. Console at https://app.o10.io/tune.
Three steps from examples to a shareable receipt.
Provide your prompt and 5–20 real input/output examples.
o10 tests prompt variants across cheaper candidate models, scoring quality on a held-out split of YOUR examples — results are never overstated.
You get a tuned prompt, a recommended model, and a shareable receipt: before/after quality, cost per 1K calls, savings %.
Sample card — labeled ILLUSTRATIVE. Not a live customer result.
BEFORE: claude-opus @ $41.20 / 1K calls · quality 0.92 → AFTER: glm-class @ $4.85 / 1K calls · quality 0.95 · savings 88%. ILLUSTRATIVE / SAMPLE.
| Before | After | |
|---|---|---|
| Model class | claude-opus | glm-class |
| Cost / 1K calls | $41.20 | $4.85 |
| Quality (held-out) | 0.92 | 0.95 |
| Savings | — | 88% ILLUSTRATIVE |
Prompt-optimization tools often charge per successful run and bill the optimization's LLM usage on top.
Recommendation-layer tools often charge for routing and/or bill optimization compute separately (Not Diamond's public pricing lists ~$0.05 per million tokens routed as of 2026-07; a fixed ~$20-per-success optimization fee is not listed on their public pricing page — needs operator confirmation before citing). o10 Tune bundles the optimization loop on low-cost routed supply, so successful tunes are free and failures never charge.
o10 runs the loop on its own low-cost routed supply, so tuning is free — failures never charge either.
Same Tune loop across console, API, Slack, and DeepShell.
Console: https://app.o10.io/tune. API: POST /v1/optimizations. Slack: @o10 tune. DeepShell: "make this task cheaper" — see /deepshell.
Sign up at https://app.o10.io/signup — free route key, no card.
Go to https://app.o10.io/tune with a prompt and 5–20 real examples.
Adopt the tuned prompt + recommended model; keep the before/after receipt.
Free prompt optimization. Submit a prompt and 5–20 real examples; o10 searches cheaper models and prompt variants and returns a tuned prompt + recommended model with a before/after receipt. Quality is measured on a held-out split of your examples. Successful tunes are free; failures never charge.
Successful tunes are free; failures never charge. o10 runs the optimization loop on its own low-cost routed supply, so you are not billed separately for the optimization's LLM usage.
Quality is measured on a held-out split of your own examples — results are never overstated. You get a before/after receipt: quality, cost per 1K calls, and savings %.
Console (https://app.o10.io/tune), API (POST /v1/optimizations), Slack (@o10 tune), and DeepShell ("make this task cheaper").
Not Diamond is a recommendation layer — it tells you which model to call; you typically pay your own provider bills, and public pricing lists ~$0.05 per million tokens routed (verify at notdiamond.ai/pricing). o10 is an in-path gateway that serves tokens, enforces the quality floor and budget, and issues savings receipts. Recommendation-layer tools often charge for routing and/or bill optimization compute separately (Not Diamond's public pricing lists ~$0.05 per million tokens routed as of 2026-07; a fixed ~$20-per-success optimization fee is not listed on their public pricing page — needs operator confirmation before citing). o10 Tune bundles the optimization loop on low-cost routed supply, so successful tunes are free and failures never charge.
Paste a week of traffic. Get the number that books the audit.
See what you're overpaying →