AI & model limits
Context windows, output, uploads, and API throughput.
Every displayed current value links to the official source.
Source-backed developer reference
Know the limit before you hit it.
Live limits, quotas, errors and changes across AI, cloud and developer platforms.
170 current observationsStructured and reusable
89 official sourcesVisible beside claims
152 reviewed lookupsNo keyword-variant sprawl
Browse the database
Start with the technology layer
Each category connects platform hubs, specific constraints, errors, comparisons, and tools.
Recently verified
Current values, source attached
These are data records rendered across lookup pages, comparisons, errors, and tools—not copied facts in separate articles.
02ChatGPT · General per-file hard limitApplies to files uploaded to a GPT or ChatGPT conversation; more restrictive type-specific limits can apply first.512 MBOpenAI
03Vercel · Function extended maximum duration (Beta)Pro and Enterprise1,800 seconds (30 minutes)Vercel
05AWS Lambda · Function timeoutA function-level configuration ceiling; the configured timeout may be lower. Functions on AWS Lambda Managed Instances may set up to 5,400 seconds (90 minutes) for asynchronous and event-source-mapping invocations.900 seconds (15 minutes)Amazon Web Services
Use the data
Free tools for the decision in front of you
Deterministic calculators use the same reviewed observations as the reference pages.
Change intelligence
Keep the old value. Record the new one.
HyperObserve uses append-only observations, so verified history is preserved instead of silently overwritten.
2026-09-17
Anthropic withdrew the published Pro message estimate
The Pro plan help article now says message capacity varies with message length, files, conversation length, model and feature; no per-five-hour count is published. Detection date recorded; no effective date given.
Built for verification
Official docs are authoritative. HyperObserve makes them comparable.
Direct answer + normalized structure + verification date + visible source + related error + useful next step.