API Uptime Scorecard: March 2026 — 42 Services Ranked by Reliability

We monitor 42 major APIs every 5 minutes from API Status Check. Not surveys. Not self-reported numbers. Raw status data pulled from official status pages, logged continuously.

Here's what March 2026 looked like across cloud infrastructure, AI, payments, developer tools, and more.

The March 2026 Scorecard

We classified each 5-minute check as operational (fully up), degraded (partial issues reported), or down (major outage). Here's every service ranked by operational uptime percentage.

🟢 99%+ Uptime — The Reliable Tier

These services stayed operational through virtually all of March:

• Canva — 100% (0 incidents)
• Figma — 100% (0 incidents)
• Jira — 100% (0 incidents)
• Robinhood — 100% (0 incidents)
• PlanetScale — 99.95% (1 incident)
• Canvas LMS — 99.78% (1 incident)
• Mailgun — 99.68% (1 incident)
• Discord — 99.62% (3 incidents)
• DeepSeek — 99.52% (3 incidents)
• Sentry — 99.57% (1 incident)
• Datadog — 99.46% (2 incidents)
• Grammarly — 99.30% (1 incident)
• Cursor — 99.20% (1 incident)
• Render — 99.03% (3 incidents)

The surprise here: DeepSeek at 99.52% outperformed every other AI service we track. While the AI category overall struggled (more on that below), DeepSeek quietly maintained near-perfect uptime through the entire month.

🟡 90-99% Uptime — The "Mostly Fine" Tier

These services experienced noticeable degradation windows but avoided prolonged outages:

• Plaid — 98.93% (2 incidents)
• Bitbucket — 98.71% (2 incidents)
• Supabase — 98.66% (6 incidents)
• Square — 98.66% (2 incidents)
• New Relic — 98.39% (2 incidents)
• Clerk — 98.29% (3 incidents)
• Confluence — 97.16% (2 incidents)
• Epic Games — 96.19% (2 incidents)
• Fortnite — 96.03% (2 incidents)
• Fly.io — 95.86% (9 incidents)
• DigitalOcean — 95.50% (6 incidents)
• Shopify — 95.43% (9 incidents)
• Netlify — 94.90% (11 incidents)
• GitHub — 94.64% (11 incidents)
• Anthropic — 92.97% (22 incidents)
• LaunchDarkly — 92.75% (2 incidents)
• Zoom — 92.70% (10 incidents)
• Linear — 91.85% (6 incidents)

Anthropic stands out — not because 92.97% is the worst uptime in this tier, but because of the frequency: 22 separate incidents in 20 days. That's more than once per day on average. Each individual incident was short-lived, but the pattern suggests systemic instability in their infrastructure.

🔴 Below 90% Uptime — The Troubled Tier

These services spent significant portions of March in a degraded state:

• SendGrid — 76.69% (2 incidents)
• ChatGPT/OpenAI — 71.76% (15 incidents)
• Coinbase — 70.94% (8 incidents)
• CockroachDB — 50.65% (2 incidents)
• Vercel — 27.92% (3 incidents)
• Elasticsearch — 18.37% (1 incident)
• Twilio — 0.64% (2 incidents)
• Grafana — 0.0% (1 incident)
• Cloudflare — 0.0% (1 incident)

Important context on the bottom three: Cloudflare, Grafana, and Twilio show extremely low uptime because their status pages reported partial service degradation across individual components for extended periods. These weren't total outages — specific subsystems (like certain Cloudflare PoPs, Grafana Cloud regions, or Twilio messaging routes) showed issues while the overall service remained functional for most users. This highlights a key challenge in status page monitoring: "degraded" doesn't always mean "broken."

The AI Reliability Gap Is Real

Across the 5 AI services we track (ChatGPT, OpenAI API, Anthropic, DeepSeek, Cursor), the average operational uptime was 88.9% in March.

For comparison:

• Consumer SaaS (Canva, Figma, Grammarly): 99.7% average
• Developer Tools (GitHub, Sentry, Datadog, Jira): 97.3% average
• Payments (Plaid, Square, Robinhood): 99.2% average
• AI Services: 88.9% average

That's a 10-point reliability gap between AI services and traditional SaaS. If you're building on AI APIs, your application inherits this unreliability. Planning for fallbacks isn't optional — it's architectural necessity.

Anthropic: Frequent But Brief

Anthropic logged 22 incidents in 20 days — the highest incident frequency of any service we track. But the pattern is distinctive: most incidents were brief degradation windows lasting 15-30 minutes. The service recovered quickly each time, suggesting aggressive auto-recovery mechanisms but underlying instability that triggers them frequently.

ChatGPT/OpenAI: Long Degradation Windows

ChatGPT and the OpenAI API (tracked separately but showing identical patterns since they share infrastructure) logged 15 incidents with 71.76% uptime. Unlike Anthropic's frequent-but-brief pattern, OpenAI's incidents tended to be longer-lasting degradation windows. When something went wrong, it stayed wrong for hours.

Infrastructure Services: The Hidden Risk

The biggest surprise in our March data isn't the AI services — it's the infrastructure layer.

Vercel (27.92% operational), CockroachDB (50.65%), and Elasticsearch (18.37%) all showed extended degradation periods. These are services that other applications depend on silently. When your database or hosting platform has a bad month, every service built on top of it inherits the problem.

This is exactly why third-party dependency monitoring matters. Your own application might be bug-free, but if your infrastructure provider is having a rough month, your users feel it.

What Perfect Uptime Actually Looks Like

Only 4 services achieved 100% operational uptime in March: Canva, Figma, Jira, and Robinhood.

What do they have in common?

1. Mature infrastructure — these are established services with years of production hardening
2. Focused product scope — each does one thing well rather than operating a sprawling platform
3. Conservative deployment patterns — fewer changes mean fewer opportunities for breakage

The common thread isn't luck — it's engineering discipline and infrastructure maturity. AI services, by contrast, are iterating rapidly on models, scaling infrastructure aggressively, and managing fundamentally new workloads. Reliability is the tax on speed.

Methodology

• Monitoring period: March 1-20, 2026
• Check interval: Every 5 minutes
• Data source: Official status pages via automated polling
• Classification: operational (all systems normal), degraded (partial issues), down (major outage)
• Services tracked: 42 across AI, cloud infrastructure, developer tools, payments, communication, and consumer SaaS
• Total data points: ~78,000 status checks

All data is collected and displayed in real-time on API Status Check. Individual service dashboards show response time trends, outage history, and current status.

Why This Matters for Your Stack

If you're building on third-party APIs (and in 2026, everyone is), reliability data should inform your architecture:

1. AI integrations need fallback logic. With sub-90% uptime across AI services, your app needs graceful degradation paths. Consider multi-provider strategies (Anthropic + OpenAI + DeepSeek).

2. Monitor your dependencies, not just your code. Traditional APM catches bugs in your code. Dependency monitoring catches when their infrastructure breaks your users' experience. API Status Check tracks 100+ services in real-time.

3. "Five nines" is a myth for most services. Only 4 out of 42 services achieved 100% uptime in a 20-day window. Plan for failure, not perfection.

4. Set up alerts before you need them. When your payment processor or AI provider goes down, you want to know immediately — not when users start complaining. Alert Pro delivers instant notifications for the services your team depends on.

Tools We Recommend for API Reliability

Based on monitoring 42+ services daily, here are the tools our team uses and recommends:

• Better Stack — The monitoring platform we trust for uptime checks, incident management, and status pages. When your dependencies go down, Better Stack tells you before your users do. If the 88.9% AI uptime stat above scares you, this is how you stay ahead of it.

• 1Password — API key management done right. When an outage forces you to rotate credentials (we saw 15 OpenAI incidents in 20 days), 1Password keeps your team's keys, tokens, and secrets organized and rotatable without Slack DMs or shared spreadsheets.

• Optery — Your engineers' personal data shows up on data broker sites. If a breached service exposes team email addresses (Coinbase at 70.9% uptime had multiple security-adjacent incidents), Optery removes your data from 380+ broker sites automatically.

---

We publish uptime scorecards regularly based on real monitoring data from API Status Check. Follow our blog for monthly updates, or monitor your dependencies in real-time.