What would your AI API bill cost on every other model?

What the $99 audit gives you

• Where your money goes: monthly cost per model and tier, cache hit rate, tokens per request.
• Every alternative, priced on your exact volume: all 71 models in our table, cheapest first, with the comparable ones marked (context window, image input and tool calling at least as good on paper). Every saving from switching carries the line: Price difference only; test quality and latency before switching.
• Same model, cheaper route: batch API, flex and off-peak tiers, prompt caching and long-context thresholds, in dollars a month.
• Recommended actions ranked by dollars a month, deprecated models you still call, and announced price changes that hit your bill.
• Files: an HTML report (save it as PDF from the page), a CSV of every row, and an email copy.

See a full sample report (from the example file, made-up volumes) · sample CSV

Monitoring, after the audit: $29/month

Prices move, and our price change log records every change we see. With monitoring we check your models against our table every hour and email you when a provider changes a price on one you use, with your new monthly total, and keep your audit current. Cancel anytime from your report page.

How to get your usage export

• OpenAI: the Usage page in the platform dashboard exports a CSV, or the Usage API (/v1/organization/usage/completions) returns JSON. Both work as is: model, input, cached input and output tokens, request counts, batch flag.
• Anthropic: the Usage page in the Claude Console exports a CSV, or the Admin API usage report (/v1/organizations/usage_report/messages) returns JSON with uncached, cache-write, cache-read and output tokens per model.
• Google: a Cloud Billing export (CSV) with SKU descriptions and usage amounts; we read the token SKUs.
• Anything else: a CSV with columns model, input_tokens, output_tokens, and optionally cached_input_tokens and requests. One row per model per month is enough.

Privacy

Send usage numbers only: model names and token counts. Never send prompts, outputs or API keys; we don't need them. We store the file you check privately so the paid report can be generated from it, and delete it on request (support@tensorust.com). The price data itself stays free and open: every model's prices, the free cost calculator and the JSON (CC BY 4.0) are not going anywhere.

What it is not

A price comparison, not a quality benchmark. Every number is list price times your tokens. Your invoice can differ by negotiated discounts, credits, regional uplifts and tool fees, and the same text tokenizes differently on other providers. Test quality and latency on your own prompts before you switch anything.