AI Tool Uptime Leaderboard

Compute uptime % over the past N days for each major AI provider (Anthropic, OpenAI, Google AI, Mistral, Cohere, ...) by pulling incident history directly from their public status pages. Outputs a ranked leaderboard with incident counts, major/minor breakdown, and per-component downtime - everything you need to compare reliability across AI vendors at a glance.

Part of Ian Mu's 100-actor portfolio at github.com/ianymu. Inspired by the post-Stop self-verification flow at github.com/ianymu/claude-verify-before-stop.

What does AI Tool Uptime Leaderboard do?

Given a list of providers and a time window (default 30 days), the Actor fetches each provider's status-page incident history. For Atlassian Statuspage-powered sites (most AI providers including Anthropic, OpenAI, Mistral, Cohere) it uses the official /api/v2/incidents.json endpoint and falls back to scraping the public /history HTML if the JSON endpoint is empty. For Google AI it uses the dedicated incidents.json Google Cloud status feed.

For each incident in the window, the Actor computes an impact-weighted duration (major = 1.0, minor = 0.5, maintenance = 0.0), sums weighted downtime, and converts that into a final uptime %. Components affected by each incident are aggregated into a per-component breakdown so you can see, for example, that the API was down for 12 minutes and the dashboard for 47.

Apify platform advantages: schedule it weekly, hit the dataset via API, integrate with Slack/webhooks, retry on transient failures - all included.

Why use AI Tool Uptime Leaderboard?

• Vendor evaluation: Compare Anthropic vs OpenAI vs Google AI reliability over the same window with the same scoring rules.
• Trust posture: Publish your own SLO dashboard fed by this Actor's dataset.
• Customer reporting: For products that depend on multiple AI providers, attach a monthly uptime report to your status page.
• Build-vs-buy: Quantify the reliability tax before locking into a vendor.

How to use AI Tool Uptime Leaderboard

1. Open the Input tab.
2. Accept the default providers (5 well-known AI vendors) or paste your own list.
3. Set daysBack (default 30; supports 1 - 365).
4. Hit Start. Runtime is short - just incident fetches, no full crawls.
5. Inspect the Output tab. Rows are ranked by uptime %, highest first.
6. Schedule it monthly for a recurring reliability report.

Input

Field	Type	Default	Description
providers	array	5 well-known AI vendors	Each item: { name, statusUrl, kind }. kind is statuspage or gcp-json.
daysBack	int	30	Time window for uptime computation. Range 1-365.

Output

Each dataset row = one provider (ranked):

{
  "rank": 1,
  "provider": "Anthropic",
  "statusUrl": "https://status.anthropic.com",
  "windowDays": 30,
  "uptimePct": 99.847,
  "weightedDowntimeMinutes": 66.0,
  "incidentCount": 4,
  "majorIncidents": 1,
  "minorIncidents": 2,
  "componentBreakdown": [
    { "component": "console.anthropic.com", "incidentCount": 2, "downtimeMinutes": 41.0 },
    { "component": "api.anthropic.com", "incidentCount": 1, "downtimeMinutes": 25.0 }
  ],
  "incidents": [ /* up to 50 most-recent incidents inside the window */ ],
  "scannedAt": "2026-05-21T10:00:00.000Z"
}

A leaderboard.json blob with the ranked summary is also written to the key-value store. You can download the dataset in JSON, HTML, CSV, or Excel.

Data table

Field	Meaning
rank	Position in the leaderboard (1 = best uptime)
provider	Vendor name
uptimePct	Impact-weighted uptime % over the window
weightedDowntimeMinutes	Total weighted downtime in minutes
incidentCount	Total incidents recorded in window
majorIncidents	Incidents marked critical/major
minorIncidents	Incidents marked minor/medium
componentBreakdown	Per-component downtime totals
incidents	Up to 50 raw incidents with timestamps and impact

Pricing / Cost estimation

A typical run (5 providers, 30 days) is just a handful of HTTP calls and runs in under a minute - well under one compute unit on Apify Free for weekly or monthly schedules.

Tips and advanced options

• Custom providers: append any Statuspage-powered service to the providers array; the statuspage kind covers anything served by Atlassian Statuspage.
• Longer windows: set daysBack up to 365 for annual SLO reports.
• Impact weighting: tweak impactWeight() in src/main.js if you want stricter or more lenient scoring (e.g., count minor incidents as 1.0).
• GCP scope: Google AI uses the GCP-wide feed; the Actor lightly weights non-AI services lower, but if you want strict Vertex/Gemini-only filtering, edit the looksAI regex.

FAQ, disclaimers, and support

• Public data only. All incident data comes from public status pages and JSON endpoints; no scraping of authenticated content.
• HTML fallback may underestimate duration. When the JSON API is unavailable, the HTML fallback assumes 60 minutes per incident; this is conservative but rough.
• GCP feed is broad. Google's incidents.json is for the entire cloud; AI-related services are detected via component name heuristic.
• Status pages can lag reality. Vendor incidents are sometimes published with delay; your number is as good as the vendor's transparency.
• Issues / feature requests: open an issue at github.com/ianymu.
• Verification methodology: developed with the post-Stop verification flow at github.com/ianymu/claude-verify-before-stop.

Author: Ian Mu - One-person AI company experiments at github.com/ianymu.