Figures on this page were verified 31 August 2026 against the providers' own documentation. Pricing, context windows and rate limits change without notice, so confirm any number against the provider before you rely on it. Tell us if something here is out of date.
Paste any LLM API error below and this tool tells you what it means, whether it is your fault, and whether retrying will help. It covers every error in our reference across OpenAI and Anthropic. Everything runs in your browser, so you can safely paste a real production stack trace.
Try one: insufficient_quota 529 overloaded context length stream closed clear
Nothing you paste leaves your browser. There is no server call, no logging and no storage.
How to read an API error in ten seconds
Two fields settle almost every case, and neither of them is the human-readable message.
- The HTTP status. A 4xx means your request and will fail identically until you change something. A 5xx means the provider and is usually worth retrying. This is the one signal that means the same thing on both APIs.
- The
typefield in the JSON body. The message text is written for humans and changes without notice. The type is the stable identifier your code should branch on.
The single exception that breaks the status-code rule is insufficient_quota, which arrives as a 429 like an ordinary rate limit but is a billing failure that no amount of backoff will clear. That one row causes more wasted engineering time than any other error in the list, which is why the decoder above suppresses the rate-limit match whenever a quota error is present.
What this tool does not do
It matches text patterns against a curated reference. It does not call an API, does not see your account, and cannot tell you why your specific request failed. If your error is not recognised, the status code and the type field are still the fastest route to an answer, and the full error reference covers the reasoning behind each verdict.
Provider error strings change. If you hit something this decoder misses or gets wrong, tell us and we will add it.
Related tools
Token volume drives both rate limits and cost. The API cost calculator prices a workload across every major model, and the context budget planner shows how much of a model window your fixed overhead consumes and at which turn a conversation breaks. Both run in the browser too.
Frequently asked questions
Does the decoder send my error anywhere?
No. The pattern matching runs entirely in your browser, so you can paste a real production stack trace without it leaving the machine. There is no server call, no logging and no storage.
What should I check if my error is not recognised?
The HTTP status code and the type field in the JSON body. The status tells you whether your request or the provider is at fault, and the type is the stable identifier to branch on. The human-readable message changes without notice and should never be matched in code.
Why does a 429 sometimes show two results?
Because 429 covers two unrelated conditions. rate_limit_exceeded is a pacing problem you should retry with backoff; insufficient_quota is a billing problem that retrying will never clear. When the body names a quota error the decoder suppresses the rate-limit match.