--- title: "OpenAI cost tracking and savings · Audon" description: "Track OpenAI API costs by project, model and line item with a read-only Admin key, and find savings in service tier, model choice, caching and output length." url: https://audon.ai/integrations/openai last_updated: 2026-11-02 --- # OpenAI cost tracking and savings. Audon reads your OpenAI organization’s Costs and Usage APIs with an Admin key, used read-only. It reconciles daily cost by project and line item, sizes savings levers on your own token data, and measures each change from billing. [Start free auditFree](https://audon.ai/sign-up?intent=spend) [About Agent Spend Optimization](https://audon.ai/product/agent-spend-optimization) ## What Audon reads. And what it never reads. Reads - Daily organization costs by project and line item - Daily token usage by model and project Never reads - Prompts or completions - Files or vector stores content - End-user data - Task outcomes ## The key to create. Used read-only, encrypted, never shown again. **Credential** Organization Admin API key **Where to create it** platform.openai.com → Organization settings → Admin keys. Only organization owners can create admin keys. **Provider documentation** [developers.openai.com/api/reference/resources/admin/subresources/organization/subresources/usage/methods/costs](https://developers.openai.com/api/reference/resources/admin/subresources/organization/subresources/usage/methods/costs) **Step by step** [Provider setup in the docs](https://audon.ai/docs/integrations#openai) ## Savings levers on OpenAI data. Sized on your own spend, each with the exact fix. - [Service tier](https://audon.ai/glossary/service-tier) Pay the fast or priority surcharge only where someone waits; run work nobody waits on at flex or batch rates. - [Regional processing](https://audon.ai/glossary/regional-processing) Drop the regional or US-only processing surcharge where no contract requires it. - [Model choice](https://audon.ai/glossary/model-choice) Move to a newer, cheaper version of the same model, or route simpler requests to a cheaper model. - [Tests and evals](https://audon.ai/glossary/tests-and-evals) Run tests, evals and CI on a cheaper model, a smaller sample or a nightly schedule. - [Context size](https://audon.ai/glossary/context-size) Stop resending growing history and stay under long-context price thresholds. - [Output and reasoning](https://audon.ai/glossary/output-and-reasoning) Cap outputs and set reasoning effort per task, so you pay for the answer, not the padding. - [Embeddings](https://audon.ai/glossary/embeddings) Embed each document once (skip unchanged content) and run bulk embedding at batch rates. - [Prompt caching](https://audon.ai/glossary/prompt-caching) Serve repeated prompt prefixes from cache, and stop paying for cache writes that are never read. - [Storage and hosted tools](https://audon.ai/glossary/storage-and-hosted-tools) Delete stored files, vector stores and caches nobody uses, and cap paid tool calls. A lever writes a potential saving. A change you mark as done is measured from billing; an approved experiment makes it a verified saving. ## Good to know. - Project keys can’t read organization costs: the key must be an organization Admin key, created by an organization owner. - OpenAI reports cost by line item, so the model is derived from the line-item label where possible. - ChatGPT plans have no billing API. Add them under Settings → Subscriptions, or let Codex send its own usage. ## Questions about OpenAI. **What does Audon need to connect OpenAI?** Organization Admin API key. platform.openai.com → Organization settings → Admin keys. Only organization owners can create admin keys. **What does Audon read from OpenAI?** Daily organization costs by project and line item and Daily token usage by model and project. **Does Audon read prompts, responses or code?** No. Audon doesn’t read: prompts or completions, files or vector stores content, end-user data and task outcomes. **Which savings levers use OpenAI data?** Service tier, regional processing, model choice, tests and evals, context size, output and reasoning, embeddings, prompt caching and storage and hosted tools. Each is sized on your own data with one stated assumption, and comes with the exact fix. **How do I revoke access?** Disconnect in Integrations and the encrypted credential is destroyed immediately. Revoke it at the provider as well, so it stops working there too. Last updated 2 November 2026 ## Other integrations. - [Anthropic (Claude API) Cost and usage reports by workspace and model, plus Claude Code per person, with one Admin key. What we read](https://audon.ai/integrations/anthropic) - [Claude EnterpriseNew Per-person activity, idle seats, and daily cost and tokens by product and model. What we read](https://audon.ai/integrations/claude-enterprise) - [Claude CodeNew Tokens and cost per person and model, from the Admin report or Claude Code’s own telemetry. What we read](https://audon.ai/integrations/claude-code) - [CodexNew Tokens per day and model from Codex’s own OpenTelemetry export, with no key stored. What we read](https://audon.ai/integrations/codex) - [xAINew Daily team spend by model and endpoint, read with a team-scoped management key. What we read](https://audon.ai/integrations/xai) - [CursorNew Members, usage per person and model and on-demand spend, or the usage export on individual plans. What we read](https://audon.ai/integrations/cursor) - [DevinNew Daily ACUs by product and per person, priced with your own price per ACU. What we read](https://audon.ai/integrations/devin) ## Smaller AI bills. Same results, proven. [Start free auditFree](https://audon.ai/sign-up?intent=spend) [See a sample report](https://audon.ai/sample-report)