BrokenAI

An independent reliability ledger for the major model providers. Built because their status pages weren't enough.

Pages
ScoreboardAbout & scoringSLA compliance30-day report
Providers
ClaudeOpenAIMistralGrokDeepSeekKimiCohere
Integrate
Status badgellms.txtSitemaprobots.txt
© 2026 BrokenAI · Independent · Not affiliated with any provider · See a number that looks wrong? @threatner_v2.1.0
BrokenAI
30-day reportSLAEmbedAbout
30-day reliability report

How did the major AI providers behave this month?

Numbers come straight from each provider's official Statuspage, recomputed on every visit. The greener cell on a row is the provider that came out ahead; ties get no tint. Hover a row for the metric's definition.

Metric
Claude
OpenAI
Mistral
Grok
DeepSeek
Kimi
Cohere
  1. Reliability
    1. Composite reliability score
      19.1
      45.1
      82.4
      93.2
      100.0
      30.2
      100.0
    2. Uptime
      66.283%
      98.546%
      99.904%
      99.982%
      100.000%
      95.168%
      100.000%
    3. Total downtime
      242h 46m
      10h 28m
      41m
      8m
      0m
      34h 47m
      0m
    4. Active right now
      0
      2
      0
      0
      0
      0
      0
  2. Incidents
    1. Total incidents
      48
      28
      22
      2
      0
      36
      0
    2. Major + critical
      14
      9
      1
      1
      0
      36
      0
    3. Minor
      31
      16
      21
      1
      0
      0
      0
    4. Longest single incident
      235h 1m
      238h 33m
      7h 15m
      ▲1h 3m
      —
      34h 43m
      —
    5. Longest clean streak
      2d 23hJul 10–13
      1d 10hJun 29 – Jul 1
      14d 7hJul 7–22
      14d 7hJul 7–22
      30dJun 22 – Jul 22
      9d 19hJul 12–22
      30dJun 22 – Jul 22
  3. Time to fix
    1. Median fix time
      1h 8m
      2h 60m
      ▲5m
      26m
      —
      16h 40m
      —
    2. On a bad day
      6h 32m
      72h 3m
      4h 4m
      ▲1h 3m
      —
      31h 43m
      —
    3. Closed vs still open
      46 closedall wrapped up
      26 closed2 still open
      22 closedall wrapped up
      2 closedall wrapped up
      0 closedall wrapped up
      36 closedall wrapped up
      0 closedall wrapped up
  4. Patterns
    1. What broke most
      Claude Code38 incidents · 154h 51m attributed
      Conversations14 incidents · 82h 16m attributed
      Workflows API5 incidents · 1h 25m attributed
      API2 incidents · 1h 28m attributed
      —
      Unspecified35 incidents · 617h 58m attributed
      —
    2. Most common failure type
      Elevated errors41 of 48
      Elevated errors12 of 28
      Degraded20 of 22
      Outage1 of 2
      —
      Other36 of 36
      —
    3. Busiest day of week
      Tuesday13 incidents on this day
      Tuesday7 incidents on this day
      Thursday7 incidents on this day
      Too few to call
      Too few to call
      Saturday24 incidents on this day
      Too few to call
    4. Busiest hour of day
      2 PM UTC7 incidents started here
      10 PM UTC3 incidents started here
      7 AM UTC6 incidents started here
      Too few to call
      Too few to call
      1 AM UTC3 incidents started here
      Too few to call
What this report doesn't cover
  • No synthetic API probing — we only see what each provider publishes on their Statuspage. Quietly degraded responses that never get acknowledged on the status page are invisible here.
  • No regional breakdown — Statuspage feeds don't reliably expose region-level impact, so a US-only outage and a global one count the same.
  • No request-volume weighting — “1 incident” means one acknowledged incident, regardless of how many calls it actually affected.