Sulat.com / Writing on AIThursday, November 19, 2026
Artificial Intelligence/18 November 2026

OpenAI’s API dashboard now shows rate-limited requests and errors by model, project or key

OpenAI Developers says Analytics in the API Platform now tracks usage and spend, rate limits and rate-limited requests, and errors by model, project or API key. Opening it takes an organisation owner role or the Usage Dashboard permission.

A white 3D OpenAI logo floats in front of a glass monitor wall showing line charts, red spike markers, bar graphs and a green gauge on a dark navy background.

At a glance

  • OpenAI Developers says Analytics in the API usage dashboard now covers usage and spend, rate limits and rate-limited requests, and errors by model, project or API key.
  • In OpenAI's demo video, Analytics lists Overview, Usage, Spend, Rate limits, API errors, Prompt caching, Safety, Codex plugins, Plugins and Service health, with filters for API sources, API keys, models and users.
  • OpenAI's error codes guide separates 429 errors by cause: rate limits and slow_down clear with pacing, while spend-limit and usage-limit errors need the limit changed before requests succeed.
  • OpenAI's help article says you need to be an organisation owner or hold the Usage Dashboard permission to open the dashboard, and that its data is shown in UTC and goes back one year.

OpenAI’s API usage dashboard now shows rate limits, rate-limited requests and errors by model, project or API key, OpenAI Developers says. Usage and spend are in the same place, under Analytics.

OpenAI Developers posted it on X at 10:51pm UTC on 17 November, just before noon on 18 November in New Zealand, along with a 23-second demo video.

What’s in Analytics

The demo opens on Analytics with a sidebar. Overview, Usage and Spend come first, then Rate limits and API errors, then Prompt caching, Safety, Codex plugins, Plugins and Service health. Filters for API sources, API keys, models and users sit below the list, next to a link to Organization analytics. Prompt caching (20 August) and Safety (23 June) were already in OpenAI’s changelog as separate launches.

The Overview page is a grid of eight tiles. They show total tokens, total requests, total spend, spend limits by project, rate limit utilisation, cache-read ratio, service health uptime and the blocked or banned request rate. The screenshots below are frames from OpenAI’s video, which OpenAI labels a prototype and fills with sample data.

OpenAI API Analytics overview: tiles for total tokens, requests, spend, spend limits, rate limit utilisation, cache-read ratio, uptime and blocked requests.
The Overview page puts spend limits, rate limit utilisation and blocked requests beside the usage totals.
Diagram: money question goes to Spend, limits question to Rate limits, failures question to API errors in OpenAI's Analytics dashboard.
Spend, Rate limits and API errors each answer one question about your API traffic.

Rate limits and rate-limited requests

The Rate limits page shows how close you’ve run to your organisation’s token ceiling, counted in tokens per minute (TPM). Current usage, peak usage and the time spent at 95% or more sit at the top. A View rate limits button opens the limits themselves, and a second chart below counts the requests that were rate limited.

OpenAI API Analytics Rate limits page: token capacity used over 30 days with current and peak usage, plus a chart of rate-limited requests.
The Rate limits page charts how much of your token capacity you used, and how many requests were rate limited.

OpenAI’s rate limits guide explains where that ceiling comes from. It applies to the organisation and to each project, never to one user, and it changes with the model. Requests per minute and tokens per minute are two of the measures, and the first one you hit trips the limit. Filters narrow the page by project, API key, model, line item, user or processing tier.

Errors by key, project or model

Errors get their own page. It plots the error count over time and lists the Top offenders below. In the demo those are API keys, ranked by their share of the errors. Per OpenAI Developers, you can explore errors by model, project or API key; the page filters cover projects, API keys and models.

OpenAI API Analytics API errors page: error count over 30 days and a Top offenders list ranking API keys by share of errors.
API errors ranks the keys behind your errors, so you can see which one to open first.

Not every 429 means you sent too much. OpenAI’s error codes guide lists several causes with different fixes. A plain rate-limit 429 needs pacing: wait at least as long as the Retry-After header says, when it’s there. A slow_down code means traffic climbed too fast. Ramp up gradually.

Some 429s are about money. organization_spend_limit_exceeded and project_spend_limit_exceeded mean you’ve reached a hard spend limit, which OpenAI added on 22 July. Retrying won’t get you back in, the guide says. Raise or remove the limit. organization_usage_limit_exceeded is different again: the organisation hit the monthly limit OpenAI assigned it, and you request a higher approved limit or contact support.

Diagram of four OpenAI 429 causes and fixes: too many requests, slow_down, spend limit reached, usage limit reached.
A 429 can be a speed limit or a money limit, and retrying only fixes the speed limit.

Filters, exports and access

OpenAI’s help article covers the controls. Usage filters by API sources, API keys, models, users and processing tier; the API sources filter splits the Agents API from direct API calls. Intervals run from a day down to a minute, and the one-minute view is where you see tokens per minute. Spend has its own filters: API sources, API keys, models and line items.

Export CSV, in the toolbar, downloads activity data. Compare dashboards shows two views side by side for the same period. Everything is in UTC and the data goes back one year. Opening the dashboard takes an organisation owner role or the Usage Dashboard permission.

Analytics lives at platform.openai.com/analytics. Organization analytics covers every project at once.

Disclosure: I’m an OpenAI, Firecrawl and Devin ambassador.


Discover more from Sulat AI & Tech News

Subscribe to get the latest posts sent to your email.