The OpenAI vs Anthropic console: where everything lives, side by side
API keys, usage, billing, rate limits, playground, logs, org structure — a side-by-side reference for the two consoles you'll switch between ten times a day if you're building with LLMs.
If you're building anything with LLMs in 2026, you have an OpenAI Platform account and an Anthropic Console account. Probably both, probably open at the same time, probably with the exact same workflow done in slightly different places.
This is a side-by-side reference for where every important thing lives in each console. Real URLs. No "leverage the power of" anywhere.
TL;DR table
| Thing | OpenAI | Anthropic |
|---|---|---|
| API keys | platform.openai.com/api-keys | console.anthropic.com/settings/keys |
| Usage / spend | platform.openai.com/usage | console.anthropic.com/settings/usage |
| Billing | platform.openai.com/account/billing | console.anthropic.com/settings/billing |
| Hard spend cap | platform.openai.com/account/limits | console.anthropic.com/settings/limits |
| Playground | platform.openai.com/playground | console.anthropic.com/workbench |
| Request logs | platform.openai.com/logs | console.anthropic.com/settings/logs |
| Org / workspaces | (under settings) | console.anthropic.com/settings/workspaces |
| Docs | platform.openai.com/docs | docs.anthropic.com |
| Status | status.openai.com | status.anthropic.com |
OK, now the detail.
API keys
OpenAI: platform.openai.com/api-keys
You create per-project keys (sk-proj-...). The old "user" keys (sk-...) still exist but project keys are the recommended path. Each key can be scoped to specific model access (/v1/chat/completions, /v1/embeddings, etc.) — useful for limiting blast radius if a key leaks.
Anthropic: console.anthropic.com/settings/keys
Anthropic keys start with sk-ant-.... They're scoped to workspaces, which are Anthropic's organizational unit. Less granular per-key permission control than OpenAI's project keys, but the workspace boundary is generally cleaner for separating "prod" from "playground."
Friction comparison: OpenAI has slightly more powerful per-key controls; Anthropic has cleaner org-level separation. Pick whichever matches how your team is already structured.
Usage & cost dashboards
OpenAI: platform.openai.com/usage
Updates in roughly 5-minute intervals (sometimes slower). Broken down by model and by day. You can filter to a specific project. The cost numbers here are authoritative — what the billing page actually charges you for.
Anthropic: console.anthropic.com/settings/usage
Updates roughly daily. Broken down by model and by workspace. Smoother UI than OpenAI's, slightly less granular timeline.
If you're cost-anxious (you should be), the move is the same on both: set a hard spend cap, set an email alert at 50% and 80%, then check the usage dashboard once a week.
The "cap" page is the most important one you've never opened
OpenAI: platform.openai.com/account/limits
Anthropic: console.anthropic.com/settings/limits
Set a hard cap the day you create the account. Not a soft alert — a hard cap. The API will start returning errors when you hit it. That is exactly what you want.
The number of "I left a loop running overnight and got a $4,000 bill" stories that show up on Twitter every week is alarming. They are essentially always preventable with a $50 cap.
Playground vs Workbench
OpenAI Playground: platform.openai.com/playground
Chat-mode and Completions-mode UIs. Plugs into Assistants. Has function-calling UI. Used to be more featureful than it is now; OpenAI has been quietly de-emphasizing it in favor of Assistants and the API directly.
Anthropic Workbench: console.anthropic.com/workbench
Anthropic's playground. Has explicit prompt-versioning and a "save as" workflow. Better than OpenAI's for actually iterating on a system prompt and saving variants. Worse if you want to compare across models.
Both are fine; neither replaces a real prompt-eval setup (e.g. Braintrust, Promptfoo, or just a Jupyter notebook with eval harness).
Logs
OpenAI: platform.openai.com/logs
Per-request log of every API call. Filter by model, by status, by latency. Showed up sometime in 2024 and it's underrated — most people only discover it after a customer reports a bug they can't reproduce.
Anthropic: console.anthropic.com/settings/logs
Similar surface. Per-request. Slightly less filterable than OpenAI's last time I checked, but ships request/response payloads, which is what you actually need for debugging.
Use both. The logs page is where "the model gave a weird answer" turns into "I can see exactly what we sent and exactly what it returned."
Org / workspace structure
This is where the two diverge.
OpenAI uses projects as the primary boundary. A project has its own keys, its own usage, its own members. You can have multiple projects in one org.
Anthropic uses workspaces. Conceptually similar to projects but with slightly different ergonomics — each workspace has its own spend limit, its own keys, and a separate "workspace admin" role.
For a solo developer, neither matters much. For a team, the choice of how to slice projects/workspaces has real consequences on billing reporting. Get it right early.
Status pages
OpenAI: status.openai.com
Anthropic: status.anthropic.com
Both are honest about outages, which is more than you can say for some providers. Subscribe to RSS or email alerts on whichever you use.
A few honest opinions
I've used both consoles roughly every working day for two years. A few patterns that aren't on either docs page:
OpenAI's UI is denser; Anthropic's is calmer. Both are fine. Anthropic's Workbench is genuinely the nicer prompt-iteration surface; OpenAI's logs page is more powerful.
OpenAI moves things around more. The path to "Usage" has shifted at least three times in two years. Anthropic's URLs have stayed put.
Both have a "hidden" billing alert toggle. Turn it on. The first time it saves you money, it pays back twice over.
If you're routing across providers, check OpenRouter or the Vercel AI Gateway before you build your own router. They handle most of the API normalization plus fallback / failover, which gets gnarly quickly if you DIY.
When you're switching between the two ten times a day
The actual annoyance isn't finding any single page once. It's finding the same page in the other console after you just looked at it in this one. ⌘K, type "openai keys" or "anthropic keys", hit enter. That's basically the whole pitch.
If you keep losing dashboard pages, that's literally what devlinkspad is for. ⌘K → devlinkspad.com.
Related:
Press ⌘K. Every dashboard, one keystroke away.
Try devlinkspad — free