One API access layer, explicit price evidence, and developer guides that keep documented facts separate from unverified behavior.
Tokza is built for developers who want to evaluate multiple AI model routes through one API and compare cost before moving production traffic.
The public site compares selected Tokza prices with dated official provider list prices only when model identity, billing unit, tier, context, modality, and currency can be matched. Your authenticated account remains authoritative for current availability and final price.
Compatibility, quality, latency, data handling, reliability, and commercial terms are separate decisions. Our guides show what to verify and leave account-specific or untested behavior unclaimed.
Our core principles guide everything we build
Material price claims show their model, unit, source, review date, formula, and limitations. Unknown fields stay unknown.
Documentation can establish a published capability. Runtime, retention, security, and reliability claims require separate evidence or contract review.
Latency and reliability should be measured on the exact account-visible route, region, workload, concurrency, and failure policy.
The API uses familiar OpenAI-compatible request patterns. The documentation also identifies where compatibility needs a real fixture instead of a promise.
Compare cost on a complete workload, including input, output, caching, retries, failures, and the accepted-task denominator.
Catalog access, features, data terms, limits, and availability can differ by account and route. Recheck before production.
Use evidence first, then a reversible test
Match an exact model and billing dimension against the dated official provider source. Reject mismatched rows.
Confirm that the route is visible in your account and review the current price, features, limits, and applicable terms.
Run a capped compatibility and billing test, reconcile usage by request, and keep the existing provider as a rollback path.
Tokza publishes selected account prices and compares them with official direct-provider list prices on an exact, dated scope. The public comparison does not infer private upstream economics.
No public page should assume identical provenance, output quality, latency, context behavior, tools, caching, or data terms from a shared model label. Test the exact route you plan to use.
This About page does not make a universal retention or certification claim. Review the current product terms and contact support with your data-flow, retention, subprocessor, and compliance requirements before sending sensitive data.
Do not infer an SLA or automatic-failover entitlement from the marketing site. Measure the route, define client-side timeout and retry boundaries, and rely only on terms that apply to your account.
The familiar request format can reduce migration work, but it does not prove drop-in compatibility. Test messages, streaming, tools, structured output, errors, usage, timeouts, and cancellation before switching traffic.
Contact sales with your route, volume, region, support, security, and contract requirements. Treat any enterprise capability as unconfirmed until it appears in a written offer that applies to your account.
Compare the route, run a capped request, and verify the billed result.