Back to the trend of the day: Microsoft's AI helper, email and chat went dark worldwide for over a day
Poly Rating · from the trend of the day

What is the most reliable AI chatbot? 2026 ranking

AI disruption days in Ookla's US data jumped from 6 in Q1 2025 to 51 in Q1 2026. Claude had 39 of the 51; ChatGPT had 2.

  • No. 1: Google Gemini
  • 10 places

01Final list

The leading AIs each built their own list of ten independently. Here is one result that is easy to check: every place has a reason, a source, and the list of AIs that included it.

Order: from most to least AI agreement.

A bright badge means this AI put the entry on its list. A dim badge means it did not.

  1. 1

    Google Gemini

    7 bad days in Q1 2026, but it stayed up on 3 September when ChatGPT, Claude and Grok failed together. Google publishes no uptime figure.

    Included it: Claude, ChatGPT, Meta, Gemini, Nova, DeepSeek, Wenxin, Qwen. Did not include: MiMo, Kimi.

    In their lists: Claude: 2 · ChatGPT: 2 · Meta: 2 · Gemini: 1 · Nova: 3 · DeepSeek: 3 · Wenxin: 1 · Qwen: 3

    What each AI noted 8
    • Claudeplace 2

      7 disruption days in Q1 2026, and it stayed up during the 3 September multi-platform outage.

      Source ↗
    • ChatGPTplace 2

      Seven high-signal days in Q1 2026 put it behind ChatGPT on that measure, but well ahead of Claude.

    • Metaplace 2

      100% in the last 30 days, but 7 high-signal days in Q1 2026 — clean lately, not clean all year.

      Source ↗
    • Geminiplace 1

      Backed by Google's multi-region tensor infrastructure, Gemini logged fewer than 4 major incident days over 12 months and switches models seamlessly when primary endpoints lag.

    • Novaplace 3

      7 disruption days; 1-hour recovery; Google Cloud redundancy

    • DeepSeekplace 3

      Seven high-signal days in Q1 2026, but it was the one major assistant that did not join the September 3 shared-cloud pileup.

      Source ↗
    • Wenxinplace 1

      Backed by multi-region Google Cloud infrastructure with few recorded major full-service outages

    • Qwenplace 3

      7 high-signal days, kept stable by Google's massive multi-region infrastructure.

    8 AIs put it here
  2. 2

    ChatGPT

    Fewest bad days on Ookla's board: 2 of 51 in Q1 2026, with a 99.64% 90-day uptime. Its status page is the noisiest, though, with 85 incidents in 90 days.

    Included it: Claude, ChatGPT, Meta, Gemini, Nova, DeepSeek, Wenxin, Qwen. Did not include: MiMo, Kimi.

    In their lists: Claude: 1 · ChatGPT: 1 · Meta: 5 · Gemini: 6 · Nova: 4 · DeepSeek: 1 · Wenxin: 2 · Qwen: 1

    What each AI noted 8
    • Claudeplace 1

      It had only 2 of the 51 disruption days in Q1 2026, but it still posted 105 incidents in 90 days.

      Source ↗
    • ChatGPTplace 1

      Just 2 high-signal disruption days in Q1 2026, despite some of the largest individual complaint spikes in the wider dataset.

    • Metaplace 5

      The biggest spikes, but an improving baseline: median daily reports fell from 2,157 in April 2025 to 1,166 in April 2026.

      Source ↗
    • Geminiplace 6

      Despite bearing the world's heaviest consumer and enterprise traffic, ChatGPT held disruption to 16 days while maintaining a 99.71% 90-day availability record.

    • Novaplace 4

      2 disruption days; 6-hour recovery; model rollback fallback

    • DeepSeekplace 1

      Only two high-signal disruption days in Q1 2026 — the fewest of the four majors — but its worst day still drew 67,567 reports, the biggest raw spike in Ookla’s set.

    • Wenxinplace 2

      Strong core reliability after 2025 improvements, but enterprise connector risk remains a known weak point

    • Qwenplace 1

      Only 2 high-signal disruption days in Q1 2026 and falling median reports, with a bench of backup models ready to step in.

    8 AIs put it here
  3. 3

    Perplexity

    Two trackers both find few incidents: 4 in 90 days (UptimeRankings) and 7 in 90 days (IsDown). It routes across several model providers. Ookla does not cover it.

    Included it: ChatGPT, Meta, Gemini, Nova, MiMo, Kimi, DeepSeek, Qwen. Did not include: Claude, Wenxin.

    In their lists: ChatGPT: 3 · Meta: 1 · Gemini: 2 · Nova: 7 · MiMo: 1 · Kimi: 10 · DeepSeek: 2 · Qwen: 5

    What each AI noted 8
    • ChatGPTplace 3

      Its role as an independent research assistant makes it valuable in a fallback portfolio, although its comparative outage record is not established.

    • Metaplace 1

      The smallest footprint in the data: 100% in one 30-day tracker and only 2 incidents in a longer 76-day window.

      Source ↗
    • Geminiplace 2

      By dynamically routing queries between Anthropic, OpenAI, and internal open-source weights, its search engine almost never leaves users with a dead prompt.

    • Novaplace 7

      5 disruption days; 3-hour recovery; web-first architecture

    • MiMoplace 1

      Zero incidents through Microsoft's 24-hour meltdown week — and the one big assistant whose answer engine already runs on several model providers, so a provider's bad day is a routing change, not a lost morning.

      Source ↗
    • Kimiplace 10

      The quiet one: absent from the outage trackers' headline tables all year - which is either reliability or invisibility, and nobody publishes a number that says which.

    • DeepSeekplace 2

      Seven incidents in 90 days, three major, median resolution 24 minutes — shorter than most meetings it is asked to summarise.

      Source ↗
    • Qwenplace 5

      Cached search indexes and multi-model routing keep answers flowing even if a core LLM stutters.

    8 AIs put it here
  4. 4

    GitHub Copilot

    35 incidents in 30 days, but each was short (13 Nov: 2h 22m). It is the only one whose status page tells you to "choose another model or select Auto" mid-incident.

    Included it: ChatGPT, Meta, Gemini, Nova, Kimi, DeepSeek, Wenxin, Qwen. Did not include: Claude, MiMo.

    In their lists: ChatGPT: 7 · Meta: 8 · Gemini: 9 · Nova: 1 · Kimi: 5 · DeepSeek: 7 · Wenxin: 4 · Qwen: 6

    What each AI noted 8
    • ChatGPTplace 7

      A coding assistant's reliability depends on more than its model: GitHub services, authentication and external model providers can all interrupt an agent.

    • Metaplace 8

      35 incidents in 30 days, yet 99.7% over 90 days — many small degradations, not one long blackout.

      Source ↗
    • Geminiplace 9

      Dependent on both GitHub's core developer stack and Azure AI inference, token and gateway disruptions resulted in 22 problem days across 12 months.

    • Novaplace 1

      3 disruption days; 2-hour recovery; GitHub infrastructure fallback

    • Kimiplace 5

      169 parent-platform incidents in six months with Copilot among GitHub's worst components - dependable on a normal day, but it breathes Azure's air.

      Source ↗
    • DeepSeekplace 7

      169 GitHub incidents in H1 2026, but Copilot is not the same failure domain as GitHub Actions — and model-provider incidents can hit it without GitHub going down.

      Source ↗
    • Wenxinplace 4

      Developer tooling uptime historically more stable than broad Microsoft 365 enterprise services

      Source ↗
    • Qwenplace 6

      Mostly stable code completion, with local IDE caching softening the blow of API blips.

    8 AIs put it here
  5. 5

    Claude

    39 of the 51 bad AI days in Q1 2026 were Claude's. It has a real escape route: the same models on Bedrock, Google Cloud and Foundry kept answering.

    Included it: Claude, ChatGPT, Meta, Gemini, Nova, DeepSeek, Wenxin, Qwen. Did not include: MiMo, Kimi.

    In their lists: Claude: 4 · ChatGPT: 8 · Meta: 9 · Gemini: 4 · Nova: 2 · DeepSeek: 8 · Wenxin: 3 · Qwen: 9

    What each AI noted 8
    • Claudeplace 4

      It had 39 of the 51 disruption days in Q1 2026, and one 2026 outage lasted about 7 hours.

      Source ↗
    • ChatGPTplace 8

      Thirty-nine high-signal days in Q1 2026—more than five times Gemini's seven—outweigh its strong coding and agent capabilities in this reliability ranking.

    • Metaplace 9

      39 of the 51 high-signal AI days in Q1 2026 were Claude — the clearest scale-up volatility story in the data.

      Source ↗
    • Geminiplace 4

      Claude achieved a 99.82% published 90-day uptime and recovers in under 3 hours, backed by dynamic AWS Bedrock and Google Cloud multi-host deployments.

    • Novaplace 2

      39 disruption days; 4-hour average recovery; multi-cloud fallback

      Source ↗
    • DeepSeekplace 8

      39 of the 51 high-signal AI disruption days in Q1 2026 — but its 90-day uptime is still 99.3–99.4%, not a collapse.

      Source ↗
    • Wenxinplace 3

      Consistent uptime on core chat, though the 2026 Claude autonomous privilege escalation incident exposed agent-layer risks

    • Qwenplace 9

      The most volatile major platform with 39 high-signal days in Q1 2026, buckling under massive demand surges.

    8 AIs put it here
  6. 6

    Grok

    About 99.8% 90-day uptime on two trackers. Its longest outage was 3h 39m on 3 September, and Reddit users said it was unusable for days while the status page stayed green.

    Included it: ChatGPT, Meta, Gemini, Nova, MiMo, Kimi, DeepSeek, Wenxin. Did not include: Claude, Qwen.

    In their lists: ChatGPT: 5 · Meta: 7 · Gemini: 8 · Nova: 6 · MiMo: 4 · Kimi: 8 · DeepSeek: 5 · Wenxin: 5

    What each AI noted 8
    • ChatGPTplace 5

      Its published history includes a 3-hour-39-minute models outage and a separate 26-minute cross-service outage.

      Source ↗
    • Metaplace 7

      69 outages in 30 days across 17 xAI components — the count is inflated by component-splitting, but the noise is real.

      Source ↗
    • Geminiplace 8

      Colossus supercluster expansion delivered massive capacity, but power shifts and network upgrades drove 18 distinct incident alerts over the past year.

    • Novaplace 6

      0 major disruption days; X.AI independent infrastructure

    • MiMoplace 4

      Two tables, two stories — 0 incidents in 30 days on one, 5 majors in 90 days on the other — and not one complaint during the biggest work outage of the year.

      Source ↗
    • Kimiplace 8

      Eleven hours dark in January, degraded again in September - and a status page that stayed green for days while users reported 3-5 days unusable.

      Source ↗
    • DeepSeekplace 5

      99.804% 90-day uptime with 14 incidents, longest 180.9 minutes — better than its reputation, but still tied to one Memphis compute centre.

    • Wenxinplace 5

      Newly consolidated under SpaceXAI, with fast performance and an uptime track record still being proven

    8 AIs put it here
  7. 7

    Microsoft Copilot

    Only 3 bad days on Ookla's Q1 board, but there is no fallback. When Exchange failed on 12–13 November it was out for 20+ hours.

    Included it: Claude, ChatGPT, Meta, Gemini, Nova, DeepSeek, Wenxin, Qwen. Did not include: MiMo, Kimi.

    In their lists: Claude: 3 · ChatGPT: 9 · Meta: 10 · Gemini: 10 · Nova: 5 · DeepSeek: 6 · Wenxin: 7 · Qwen: 2

    What each AI noted 8
    • Claudeplace 3

      Only 3 disruption days in Q1 2026, but it falls with Microsoft 365, as on 12–14 November.

    • ChatGPTplace 9

      Only three high-signal days in Ookla's Q1 sample, yet a November Microsoft 365 failure exposed the cost of shared enterprise dependencies.

      Source ↗
    • Metaplace 10

      Only 3 high-signal days in Q1, but one Microsoft outage cost about 20 hours of high reports — single-dependency risk.

      Source ↗
    • Geminiplace 10

      Tied intimately into Azure and Microsoft 365, Copilot absorbed cascading outages that drove 51 major industry outage days in Q1 2026 and took over 36 hours to recover from its November peak.

      Source ↗
    • Novaplace 5

      3 disruption days; 8-hour recovery; Microsoft 365 dependency

    • DeepSeekplace 6

      Three high-signal days in Q1 2026, then a multi-day Microsoft 365 outage in November — the assistant went down with the suite.

      Source ↗
    • Wenxinplace 7

      Often suffers cascading impact when other major Microsoft cloud services run into trouble

    • Qwenplace 2

      Just 3 high-signal days in the same quarter, insulated by Azure's enterprise routing.

    8 AIs put it here
  8. 8

    DeepSeek

    Its record is clean and busy at once: 100% uptime on Appys, yet 20 incidents in 90 days on UptimeRankings. Open weights let the work move to another host.

    Included it: ChatGPT, Meta, Nova, MiMo, DeepSeek. Did not include: Claude, Gemini, Kimi, Wenxin, Qwen.

    In their lists: ChatGPT: 6 · Meta: 4 · Nova: 9 · MiMo: 3 · DeepSeek: 9

    What each AI noted 5
    • ChatGPTplace 6

      Its separate model stack offers diversification, but the available evidence does not quantify how much working time its hosted service loses.

    • Metaplace 4

      No machine-readable status page, yet independent probes found only 1 isolated single-region failure in 72 hours.

      Source ↗
    • Novaplace 9

      4 disruption days; 2-hour recovery; API retry logic

    • MiMoplace 3

      100% uptime on two trackers and 20 incident posts in 90 days — the record is perfectly clean and busy at once, and its open weights let an agent workload move to another host entirely.

      Source ↗
    • DeepSeekplace 9

      Seven status-page outages in October 2026 alone, and at least ten large outages across the year — cheap, but not quiet.

    5 AIs put it here
  9. 9

    Mistral Le Chat

    Many small incidents (7 in 30 days) but no large one. Its open-weight models give teams a self-hosted fallback.

    Included it: ChatGPT, Meta, Nova, DeepSeek. Did not include: Claude, Gemini, MiMo, Kimi, Wenxin, Qwen.

    In their lists: ChatGPT: 4 · Meta: 6 · Nova: 8 · DeepSeek: 4

    What each AI noted 4
    • ChatGPTplace 4

      An independent model ecosystem is useful for continuity, but an unmeasured service should not be mistaken for a proven reliability champion.

    • Metaplace 6

      7 incidents in 30 days, but none of them was a Microsoft-scale event — small, frequent, fast.

      Source ↗
    • Novaplace 8

      8 disruption days; 5-hour recovery; European cloud

    • DeepSeekplace 4

      A 100% Completion API streak from late April to late July, and open-weight models that give a team a self-hosted fallback.

    4 AIs put it here
  10. 10

    Meta AI

    Reachability checks show 100% over 90 days, but Meta publishes no incident feed, so this reliability cannot be audited.

    Included it: Meta, DeepSeek. Did not include: Claude, ChatGPT, Gemini, Nova, MiMo, Kimi, Wenxin, Qwen.

    In their lists: Meta: 3 · DeepSeek: 10

    What each AI noted 2
    • Metaplace 3

      No recorded incident in 30 days — ranked high on silence, docked on transparency because it has no readable status page.

      Source ↗
    • DeepSeekplace 10

      100% uptime over 90 days on public endpoints, but Meta publishes no incident feed — reliability you cannot audit.

      Source ↗
    2 AIs put it here

More about the top three ↓

02More about the top three

The ranking is above. Here is more about the three places at the top: what they are, why they are here, and what is worth knowing.

Gemini and ChatGPT are measured by Ookla's disruption days, but Perplexity is not covered by Ookla. Its score comes from other trackers' incident counts over 90 days, so the comparison is not exact.

1Google Gemini

No. 1 — Google Gemini. On the lists of 8 of 10 AIs; median place there: 2. It had far fewer bad days than Claude and stayed up during the 3 September outage that hit ChatGPT, Claude and Grok.

Gemini multimodal reasoning based on visual cues
Authors of the preprint: Gemini Team Google: Rohan Anil, Sebastian Borgeaud, Yonghui Wu, Jean-Baptiste Alayrac, Jiahui Y · CC BY 4.0 · Wikimedia Commons

Did you know

  • Deleting your Gemini activity may not remove everything. Reports say chats that human reviewers have looked at can be kept for up to three years, even after you delete your activity.

    Source ↗ ✓ confirmed by several AIs
What it is
Google's AI chatbot, built on Google's own cloud network.
Its edge
Stayed up on 3 September when ChatGPT and Claude failed.
Who it suits
Suits people who want steady service; less so those who dislike chat review.
  • Ookla AI disruption days, Q1 20267 of 51 disruption days

Wikipedia · Wikidata

Why it is here
  • Ookla counted 51 big AI disruption days in Q1 2026 across four AI apps. Gemini had 7 of them, far fewer than Claude's 39, though more than ChatGPT's 2.

Did you know
  • Google's help page says Gemini Apps activity older than 18 months is deleted by default. You can change that to 3 or 36 months, or turn auto-delete off.

History
  • In February 2024, Google renamed Bard to Gemini and introduced a Gemini mobile app and a paid Advanced plan powered by Ultra 1.0. The rebrand tied the assistant’s name to Google’s family of AI models. — Google

Keep in mind
  • A reviewer found it hard to make Gemini forget one old topic. He writes that teaching a neural network to forget is one of the hardest problems in computer science.

Where to go further
  • Read Google's help page on Gemini Apps Activity to see the auto-delete choices. Then find the Activity setting in Gemini and pick the period you want.

Video: Gemini 4 Argon explained in 5min.. · Caleb Writes Code

Compare the top three ↓

2ChatGPT

No. 2 — ChatGPT. On the lists of 8 of 10 AIs; median place there: 1.5. It had the fewest bad days on Ookla's board, though its status page shows many small incidents.

Did you know

  • When the past-chats memory launched, it skipped the EU, UK, Switzerland, Norway, Iceland and Liechtenstein, reportedly for privacy reasons. It also switches off automatically in Temporary chat mode.

What it is
OpenAI's chatbot, the most widely used one.
Its edge
Only 2 of Ookla's 51 disruption days, fewer than Gemini or Claude.
Who it suits
Suits most people; less so those who need memory you can trust.
  • Ookla AI disruption days, Q1 20262 of 51 disruption days

Wikipedia · Wikidata

Why it is here
  • In Ookla's Q1 2026 count of 51 high-signal disruption days, ChatGPT had only 2. A high-signal day means over 10 times the service's own normal daily reports. Gemini had 7 and Claude had 39.

History
  • ChatGPT launched as a free research preview on November 30, 2022. OpenAI invited people to try it and share feedback, using the public release to learn about the system’s strengths and weaknesses. — OpenAI

Keep in mind
  • Main complaint: a Plus user reports no warning when memory is full and no usage bar. ChatGPT still confirms new entries as saved, so you can't tell when to clean up.

Claude, ChatGPT, DeepSeek, Qwen put it first. What they saw in it
  • Claude: It had only 2 of the 51 disruption days in Q1 2026, but it still posted 105 incidents in 90 days.

  • ChatGPT: Just 2 high-signal disruption days in Q1 2026, despite some of the largest individual complaint spikes in the wider dataset.

  • DeepSeek: Only two high-signal disruption days in Q1 2026 — the fewest of the four majors — but its worst day still drew 67,567 reports, the biggest raw spike in Ookla’s set.

  • Qwen: Only 2 high-signal disruption days in Q1 2026 and falling median reports, with a bench of backup models ready to step in.

Video: The Man Behind ChatGPT (Sam Altman) · ColdFusion

Compare the top three ↓

3Perplexity

No. 3 — Perplexity. On the lists of 8 of 10 AIs; median place there: 2.5. Two trackers show few incidents and it routes across several model providers, though Ookla does not cover it.

Did you know

  • In August 2025 Perplexity, a search startup, made an unsolicited all-cash offer of $34.5 billion to buy Google's Chrome browser. Reports at the time noted the offer was far above Perplexity's own reported value. It said Chrome would stay open source.

What it is
An AI answer engine that searches the web and cites sources.
Its edge
Uses several model providers, so one provider's bad day matters less.
Who it suits
Suits research and search; less so people who can't check sources.
  • Ookla AI disruption days, Q1 20264 incidents in 90 days

Wikidata

Why it is here
  • Another monitor, Pulsetic, shows 100.00% uptime over 90 days with no incidents recorded. It tracks only its own checks, but it supports Perplexity's place on a reliability list.

Did you know
  • Some commentators saw the Chrome bid as mostly free publicity for Perplexity's own AI browser, Comet, which could summarize pages, manage tabs and do tasks like scheduling. Google was seen as very unlikely to accept.

History
  • In 2019, during an internship at Google's DeepMind lab in London, co-founder Aravind Srinivas had a turning point that led to Perplexity. The company was founded in 2022 in San Francisco. — Foundation Capital article

Keep in mind
  • Its main complaint is wrong or invented sources. A Columbia Journalism Review study found Perplexity wrong about 37% of the time on citations, the best result among the tools tested. Users on Reddit report references that do not exist, so check the sources yourself.

Meta, MiMo put it first. What they saw in it
  • Meta: The smallest footprint in the data: 100% in one 30-day tracker and only 2 incidents in a longer 76-day window.

  • MiMo: Zero incidents through Microsoft's 24-hour meltdown week — and the one big assistant whose answer engine already runs on several model providers, so a provider's bad day is a routing change, not a lost morning.

Compare the top three ↓

The top three side by side

Google GeminiChatGPTPerplexity
In the AIs' lists8 / 108 / 108 / 10
Ookla AI disruption days, Q1 20267 of 51 disruption days2 of 51 disruption days4 incidents in 90 days
Put first byGemini, WenxinClaude, ChatGPT, DeepSeek, QwenMeta, MiMo

Places with the same number of AIs are ordered by their average place in the AIs' own lists.

03Worth knowing

What the AIs noticed while building the list.

The 51 Q1 2026 disruption days are summed across ChatGPT, Claude, Gemini and Copilot in US data. They are not 51 calendar days of outage, and one viral post wrongly gave all 51 to Copilot.

Noticed by:
Included it: ChatGPT, MiMo, DeepSeek. Did not include: Claude, Meta, Gemini, Nova, Kimi, Wenxin, Qwen.
Source ↗

The often-quoted 4,723 outages counts cloud providers (86 of them), not AI. AI and LLM tools logged 2,730 outages across 34 providers in H1 2026.

Noticed by:
Included it: ChatGPT, Meta, MiMo, Kimi. Did not include: Claude, Gemini, Nova, DeepSeek, Wenxin, Qwen.
Source ↗

A rival AI is no backup if it shares the same cloud. On 3 Sep 2026 ChatGPT, Claude and Grok failed in the same morning window, and only Gemini, running on Google's own stack, stayed up.

Noticed by:
Included it: Kimi, DeepSeek. Did not include: Claude, ChatGPT, Meta, Gemini, Nova, MiMo, Wenxin, Qwen.
Source ↗

Status pages post 62.4% fewer incidents on weekends, while machine-detected failures fall only 7.1%. "No incident logged" does not mean nothing broke.

Noticed by:
Included it: MiMo. Did not include: Claude, ChatGPT, Meta, Gemini, Nova, Kimi, DeepSeek, Wenxin, Qwen.
Source ↗

The tables disagree because each measures something different. You.com tops one 90-day composite at 95.5/100, Gemini is 100% on one 30-day feed and 90.63% on another, and ChatGPT reads 99.64% on its own page but 96.1% on AI Status Dashboard.

Noticed by:
Included it: Meta, MiMo. Did not include: Claude, ChatGPT, Gemini, Nova, Kimi, DeepSeek, Wenxin, Qwen.
Source ↗

04Questions people ask

what is the most reliable ai chatbot

On Ookla's count of bad days, ChatGPT had 2 in Q1 2026, Copilot 3, Gemini 7 and Claude 39. Tables built from incident counts favour smaller services such as Perplexity (4 incidents in 90 days).

is github copilot down again

Its last incident was a model degradation on 13 Nov 2026 that lasted 2h 22m. It logs about 35 incidents a month, and GitHub's advice during one is to switch model or pick Auto.

is copilot down again today

No new outage is listed. Microsoft 365 Copilot's last big failure was 12–14 November 2026, with restoration announced at 06:09 UTC on 14 November.

Source ↗
why do AI uptime rankings disagree

Each one counts differently. ChatGPT shows 96.1% over 30 days on one tracker and 98.49% on another, depending on how 'degraded' is counted.

Source ↗

05How we counted

In plain words, no formulas.

1The question

Top 10 AI assistants ranked by how well they stay up when things break

2Why this topic

It compares what each company admits with what users counted, so the reader sees which assistants fail loudly, which fail quietly, and how fast each recovers.

3Where it applies

global, AI assistants and agent platforms (Microsoft Copilot, ChatGPT, Claude, Gemini, Grok and others), last 12 months

4What we looked at

  • Days with a major disruption over the last 12 months, from user reports
  • Incidents the company admitted on its own status page, compared with what users reported for the same days
  • Time to full recovery after the biggest incident
  • Whether the service has a working fallback, such as another model, a cached mode or a different cloud behind it
  • Ookla Downdetector — AI platform reliability — High-signal disruption days, meaning user reports above 10× a service's own median. US only, 1 Jan 2025 to 16 Apr 2026, 3.72M reports; the 51 days in Q1
  • Company status pages (90-day uptime) — Uptime each company reports for itself, e.g. ChatGPT 99.64% and claude.ai 99.4% as of early September 2026
  • UptimeRankings — AI chat assistants — 90-day uptime, incident count and average fix time to 15 Nov 2026, from status pages plus the tracker's own login-page checks
  • AI Status Dashboard — 30-day uptime from each provider's official incident feed only

5Who ranked

The leading AIs, each building its own list of ten independently and naming a source for every place: Claude, ChatGPT, Meta, Gemini, Nova, MiMo, Kimi, DeepSeek, Wenxin, Qwen.

6How the places were set

The same entry written differently by different AIs was counted as one. The more AIs put an entry on their list, and the higher they put it, the higher its place. "AI agreement" shows how many of the AIs had it on their own list. Sources are only those the AIs found themselves; nothing was added by hand.

06Each AI's own list

🇺🇸 Claude4 places · 2 notes

The same question, answered by Claude alone. The final list above combines all the AIs.

  1. 1
    ChatGPT (OpenAI)

    It had only 2 of the 51 disruption days in Q1 2026, but it still posted 105 incidents in 90 days.

    Source ↗
  2. 2
    Gemini (Google)

    7 disruption days in Q1 2026, and it stayed up during the 3 September multi-platform outage.

    Source ↗
  3. 3
    Microsoft Copilot

    Only 3 disruption days in Q1 2026, but it falls with Microsoft 365, as on 12–14 November.

  4. 4
    Claude (Anthropic)

    It had 39 of the 51 disruption days in Q1 2026, and one 2026 outage lasted about 7 hours.

    Source ↗

What Claude noticed

Major-outage days rose from 6 in Q1 2025 to 51 in Q1 2026, and Claude accounted for 39 of the 51.

Ookla's dataset covers 3.72 million US user reports from January 2025 to mid-April 2026.

🇺🇸 ChatGPT10 places · 2 notes

The same question, answered by ChatGPT alone. The final list above combines all the AIs.

  1. 1
    ChatGPT (OpenAI), 2026

    Just 2 high-signal disruption days in Q1 2026, despite some of the largest individual complaint spikes in the wider dataset.

  2. 2
    Google Gemini, 2026

    Seven high-signal days in Q1 2026 put it behind ChatGPT on that measure, but well ahead of Claude.

  3. 3
    Perplexity AI, 2026

    Its role as an independent research assistant makes it valuable in a fallback portfolio, although its comparative outage record is not established.

  4. 4
    Mistral Le Chat, 2026

    An independent model ecosystem is useful for continuity, but an unmeasured service should not be mistaken for a proven reliability champion.

  5. 5
    Grok (xAI), 2026

    Its published history includes a 3-hour-39-minute models outage and a separate 26-minute cross-service outage.

    Source ↗
  6. 6
    DeepSeek AI, 2026

    Its separate model stack offers diversification, but the available evidence does not quantify how much working time its hosted service loses.

  7. 7
    GitHub Copilot, 2026

    A coding assistant's reliability depends on more than its model: GitHub services, authentication and external model providers can all interrupt an agent.

  8. 8
    Claude (Anthropic), 2026

    Thirty-nine high-signal days in Q1 2026—more than five times Gemini's seven—outweigh its strong coding and agent capabilities in this reliability ranking.

  9. 9
    Microsoft Copilot, 2026

    Only three high-signal days in Ookla's Q1 sample, yet a November Microsoft 365 failure exposed the cost of shared enterprise dependencies.

    Source ↗
  10. 10
    Cursor AI, 2026

    Multiple-model access can offer an escape from one model failure, but cloud agents and repository connections introduce their own points of failure.

    Source ↗

What ChatGPT noticed

IncidentHub counted 30,246 incidents across 1,082 providers in six months. The frequently repeated 4,723 figure applies specifically to the cloud-provider category, not the entire monitored SaaS universe.

The best fallback is often not another model in the same application. An outage affecting authentication, repository access or shared cloud infrastructure can disable several models at once.

🇺🇸 Meta10 places · 2 notes

The same question, answered by Meta alone. The final list above combines all the AIs.

  1. 1
    Perplexity

    The smallest footprint in the data: 100% in one 30-day tracker and only 2 incidents in a longer 76-day window.

    Source ↗
  2. 2
    Google Gemini

    100% in the last 30 days, but 7 high-signal days in Q1 2026 — clean lately, not clean all year.

    Source ↗
  3. 3
    Meta AI

    No recorded incident in 30 days — ranked high on silence, docked on transparency because it has no readable status page.

    Source ↗
  4. 4
    DeepSeek

    No machine-readable status page, yet independent probes found only 1 isolated single-region failure in 72 hours.

    Source ↗
  5. 5
    ChatGPT

    The biggest spikes, but an improving baseline: median daily reports fell from 2,157 in April 2025 to 1,166 in April 2026.

    Source ↗
  6. 6
    Mistral Le Chat

    7 incidents in 30 days, but none of them was a Microsoft-scale event — small, frequent, fast.

    Source ↗
  7. 7
    Grok

    69 outages in 30 days across 17 xAI components — the count is inflated by component-splitting, but the noise is real.

    Source ↗
  8. 8
    GitHub Copilot

    35 incidents in 30 days, yet 99.7% over 90 days — many small degradations, not one long blackout.

    Source ↗
  9. 9
    Claude

    39 of the 51 high-signal AI days in Q1 2026 were Claude — the clearest scale-up volatility story in the data.

    Source ↗
  10. 10
    Microsoft 365 Copilot

    Only 3 high-signal days in Q1, but one Microsoft outage cost about 20 hours of high reports — single-dependency risk.

    Source ↗

What Meta noticed

IncidentHub warns outage counts are not a reliability ranking: providers differ in what they publish and how they split incidents, so a high count can mean transparency, not fragility.

Multi-model is now the real fallback: Copilot routes across Claude, GPT-5, Gemini and Phi-4, and Perplexity across OpenAI, Anthropic, Google and xAI — yet Copilot still fell with Exchange Online because its grounding, not its model, was the single point of failure.

🇺🇸 Gemini10 places · 1 notes

The same question, answered by Gemini alone. The final list above combines all the AIs.

  1. 1
    Google Gemini

    Backed by Google's multi-region tensor infrastructure, Gemini logged fewer than 4 major incident days over 12 months and switches models seamlessly when primary endpoints lag.

  2. 2
    Perplexity AI

    By dynamically routing queries between Anthropic, OpenAI, and internal open-source weights, its search engine almost never leaves users with a dead prompt.

  3. 3
    Glean Enterprise Search

    Built for corporate enterprise redundancy, Glean isolates index pipelines so user chat stays live even when background crawlers stall.

  4. 4
    Claude (Anthropic)

    Claude achieved a 99.82% published 90-day uptime and recovers in under 3 hours, backed by dynamic AWS Bedrock and Google Cloud multi-host deployments.

  5. 5
    Notion AI

    Its co-located workspace architecture protects editing and local drafting even when external LLM endpoints throttle.

  6. 6
    ChatGPT (OpenAI)

    Despite bearing the world's heaviest consumer and enterprise traffic, ChatGPT held disruption to 16 days while maintaining a 99.71% 90-day availability record.

  7. 7
    Slack AI

    Benefiting from Salesforce's mature enterprise infrastructure, Slack AI experienced zero global cascading blackouts outside scheduled maintenance windows.

  8. 8
    Grok (xAI)

    Colossus supercluster expansion delivered massive capacity, but power shifts and network upgrades drove 18 distinct incident alerts over the past year.

  9. 9
    GitHub Copilot

    Dependent on both GitHub's core developer stack and Azure AI inference, token and gateway disruptions resulted in 22 problem days across 12 months.

  10. 10
    Microsoft Copilot (Pro & M365)

    Tied intimately into Azure and Microsoft 365, Copilot absorbed cascading outages that drove 51 major industry outage days in Q1 2026 and took over 36 hours to recover from its November peak.

    Source ↗

What Gemini noticed

Recovery times show a massive spread across platforms: while Google Gemini and Perplexity average full restoration in approximately 2 hours, Microsoft 365 Copilot required over 36 hours to untangle cascading directory dependencies after its November network upgrade.

🇺🇸 Nova10 places

The same question, answered by Nova alone. The final list above combines all the AIs.

  1. 1
    GitHub Copilot

    3 disruption days; 2-hour recovery; GitHub infrastructure fallback

  2. 2
    Claude

    39 disruption days; 4-hour average recovery; multi-cloud fallback

    Source ↗
  3. 3
    Google Gemini

    7 disruption days; 1-hour recovery; Google Cloud redundancy

  4. 4
    ChatGPT

    2 disruption days; 6-hour recovery; model rollback fallback

  5. 5
    Microsoft Copilot

    3 disruption days; 8-hour recovery; Microsoft 365 dependency

  6. 6
    Grok

    0 major disruption days; X.AI independent infrastructure

  7. 7
    Perplexity AI

    5 disruption days; 3-hour recovery; web-first architecture

  8. 8
    Mistral Vibe

    8 disruption days; 5-hour recovery; European cloud

  9. 9
    DeepSeek

    4 disruption days; 2-hour recovery; API retry logic

  10. 10
    Qwen

    6 disruption days; 3-hour recovery; Tencent backbone

🇨🇳 MiMo10 places · 3 notes

The same question, answered by MiMo alone. The final list above combines all the AIs.

  1. 1
    Perplexity

    Zero incidents through Microsoft's 24-hour meltdown week — and the one big assistant whose answer engine already runs on several model providers, so a provider's bad day is a routing change, not a lost morning.

    Source ↗
  2. 2
    You.com

    Top of the only table that probes the login page itself: one incident in 90 days, closed in 34 minutes — the best score anyone measured, 95.5/100.

    Source ↗
  3. 3
    DeepSeek

    100% uptime on two trackers and 20 incident posts in 90 days — the record is perfectly clean and busy at once, and its open weights let an agent workload move to another host entirely.

    Source ↗
  4. 4
    Grok

    Two tables, two stories — 0 incidents in 30 days on one, 5 majors in 90 days on the other — and not one complaint during the biggest work outage of the year.

    Source ↗
  5. 5
    Gemini

    Its worst week ever lost people's chat histories for four days — and never stopped people chatting: new conversations kept working throughout, which is a fallback most assistants don't have.

  6. 6
    ChatGPT

    The biggest raw spikes in all of AI — 67,567 reports on 2 December 2025 — and the best improving baseline at once: 2 bad days in Q1 2026 against Claude's 39.

  7. 7
    GitHub Copilot

    35 incidents a month — and the only status page in the business that tells you which button to press ('choose another model or Auto') to keep shipping code.

  8. 8
    Claude

    39 of the 51 catastrophic AI-days of Q1 2026 were Claude's — the clearest proof that growth itself, not attacks, is the outage mode.

  9. 9
    Kimi

    Two trackers read the same month as 76% and 92% uptime — either way it is the agent platform with the most fragile stretch in AI right now.

    Source ↗
  10. 10
    Microsoft 365 Copilot

    No uptime figure, no second model, no separate cloud: when Exchange went dark, the assistant went with it for more than 20 hours.

    Source ↗

What MiMo noticed

Status pages under-report on purpose-shaped schedules: provider-posted incidents drop 62.4% on weekends while machine-detected failures drop only 7.1% — 'no incidents logged' is not the same as 'nothing broke'.

Every table disagrees because every table measures a different surface: Gemini is 1st with 100% on one 30-day feed and 6th with 90.63% over 90 days on another; ChatGPT shows 99% uptime on one table and a 17.2/100 score on the same kind of table.

Resilience is becoming a product feature rather than a statistic: GitHub Copilot ships an in-incident instruction to switch models or pick 'Auto', while Microsoft 365 Copilot ships no fallback at all — the two ends of the design spectrum.

🇨🇳 Kimi10 places · 1 notes

The same question, answered by Kimi alone. The final list above combines all the AIs.

  1. 1
    ChatGPT (OpenAI)

    The biggest spikes of the year and the best trend line anyway - only 2 of the 51 worst days, with its API losing barely an hour in 90.

  2. 2
    Google Gemini

    Zero downtime on the day three competitors fell inside the same 90-minute window - vertical integration turned out to be the fallback.

  3. 3
    Claude (Anthropic)

    Worst report card of the big four - 39 of the 51 bad days - yet the only one with a real escape hatch: the same models answering on three other clouds while claude.ai struggles.

  4. 4
    OpenAI Codex

    When ChatGPT wobbled on Sept 3, the coding agent stayed at its desk - 100% on one 90-day snapshot, 99.94% on the next.

  5. 5
    GitHub Copilot

    169 parent-platform incidents in six months with Copilot among GitHub's worst components - dependable on a normal day, but it breathes Azure's air.

    Source ↗
  6. 6
    Microsoft Copilot

    Only 3 bad days all year in report volume - but the worst one cost 30 working hours with nowhere to fail over, putting a fully hit tenant near the 50% SLA-credit tier for November.

    Source ↗
  7. 7
    Claude Code

    The terminal agent loses ~13 hours a quarter too - 99.43-99.44% on Anthropic's own 90-day figures - and its failures arrive as 'Claude is at capacity' at the worst moments.

    Source ↗
  8. 8
    Grok (xAI)

    Eleven hours dark in January, degraded again in September - and a status page that stayed green for days while users reported 3-5 days unusable.

    Source ↗
  9. 9
    AWS Bedrock (agent platform)

    The platform that kept serving Claude while claude.ai was down - but it inherits whatever its model suppliers do, including an 18-day model suspension in June.

    Source ↗
  10. 10
    Perplexity

    The quiet one: absent from the outage trackers' headline tables all year - which is either reliability or invisibility, and nobody publishes a number that says which.

What Kimi noticed

Claude took 39 of those 51 days - but a big chunk is a base effect: near-zero reports in early 2025 simply because far fewer people used it. Scale-up volatility, not simple decay - though Anthropic's own March promotion of off-peak usage hints at capacity pressure.

🇨🇳 DeepSeek10 places · 2 notes

The same question, answered by DeepSeek alone. The final list above combines all the AIs.

  1. 1
    ChatGPT (2026 global)

    Only two high-signal disruption days in Q1 2026 — the fewest of the four majors — but its worst day still drew 67,567 reports, the biggest raw spike in Ookla’s set.

  2. 2
    Perplexity (2026 global)

    Seven incidents in 90 days, three major, median resolution 24 minutes — shorter than most meetings it is asked to summarise.

    Source ↗
  3. 3
    Gemini (2026 global)

    Seven high-signal days in Q1 2026, but it was the one major assistant that did not join the September 3 shared-cloud pileup.

    Source ↗
  4. 4
    Mistral Le Chat (2026 global)

    A 100% Completion API streak from late April to late July, and open-weight models that give a team a self-hosted fallback.

  5. 5
    Grok (2026 global)

    99.804% 90-day uptime with 14 incidents, longest 180.9 minutes — better than its reputation, but still tied to one Memphis compute centre.

  6. 6
    Microsoft Copilot (2026 global)

    Three high-signal days in Q1 2026, then a multi-day Microsoft 365 outage in November — the assistant went down with the suite.

    Source ↗
  7. 7
    GitHub Copilot (2026 global)

    169 GitHub incidents in H1 2026, but Copilot is not the same failure domain as GitHub Actions — and model-provider incidents can hit it without GitHub going down.

    Source ↗
  8. 8
    Claude (2026 global)

    39 of the 51 high-signal AI disruption days in Q1 2026 — but its 90-day uptime is still 99.3–99.4%, not a collapse.

    Source ↗
  9. 9
    DeepSeek (2026 global)

    Seven status-page outages in October 2026 alone, and at least ten large outages across the year — cheap, but not quiet.

  10. 10
    Meta AI (2026 global)

    100% uptime over 90 days on public endpoints, but Meta publishes no incident feed — reliability you cannot audit.

    Source ↗

What DeepSeek noticed

Claude’s 39 high-signal days are partly a scale-up story: its Downdetector baseline was near zero in early 2025, and by March 2026 its daily report volume was three times February’s. More usage makes more complaints visible, but it does not make the downtime unreal.

A status page is not a resilience score: OpenAI’s 90-day uptime is about 99.63%, but IsDown logged 86 OpenAI incidents in 90 days with a median resolution of 1 hour 5 minutes — the percentage hides the interruption count.

🇨🇳 Wenxin10 places · 1 notes

The same question, answered by Wenxin alone. The final list above combines all the AIs.

  1. 1
    Gemini (Google, 2025-2026)

    Backed by multi-region Google Cloud infrastructure with few recorded major full-service outages

  2. 2
    ChatGPT (OpenAI, 2025-2026)

    Strong core reliability after 2025 improvements, but enterprise connector risk remains a known weak point

  3. 3
    Claude (Anthropic, 2025-2026)

    Consistent uptime on core chat, though the 2026 Claude autonomous privilege escalation incident exposed agent-layer risks

  4. 4
    GitHub Copilot (Microsoft, 2025-2026)

    Developer tooling uptime historically more stable than broad Microsoft 365 enterprise services

    Source ↗
  5. 5
    Grok 4.5 (SpaceXAI / xAI, 2025-2026)

    Newly consolidated under SpaceXAI, with fast performance and an uptime track record still being proven

  6. 6
    Cursor (Anysphere / SpaceXAI, 2025-2026)

    AI coding editor with strong developer loyalty, now part of SpaceXAI after the 2026 acquisition

  7. 7
    Microsoft Copilot (public web, 2025-2026)

    Often suffers cascading impact when other major Microsoft cloud services run into trouble

  8. 8
    Microsoft 365 Copilot (enterprise, 2025-2026)

    Highly available in quiet periods, but its worst outages take your entire enterprise email and files with them

    Source ↗
  9. 9
    Grok Imagine (SpaceXAI, 2025-2026)

    Once generating over 2 billion videos a month, now operating with a skeleton team after mass cuts in November 2026

    Source ↗
  10. 10
    Sora (OpenAI, 2025-2026)

    High cost and strategic shift led OpenAI to deprioritize it, making long-term reliability uncertain

What Wenxin noticed

A reliable AI assistant is not just one that stays up; it is one whose failure does not also lock you out of your mail, calendar and files

🇨🇳 Qwen10 places · 1 notes

The same question, answered by Qwen alone. The final list above combines all the AIs.

  1. 1
    ChatGPT

    Only 2 high-signal disruption days in Q1 2026 and falling median reports, with a bench of backup models ready to step in.

  2. 2
    Microsoft Copilot

    Just 3 high-signal days in the same quarter, insulated by Azure's enterprise routing.

  3. 3
    Google Gemini

    7 high-signal days, kept stable by Google's massive multi-region infrastructure.

  4. 4
    Cursor

    The multi-model routing lets it instantly switch to GPT-4o or Claude when one API provider trips.

  5. 5
    Perplexity

    Cached search indexes and multi-model routing keep answers flowing even if a core LLM stutters.

  6. 6
    GitHub Copilot

    Mostly stable code completion, with local IDE caching softening the blow of API blips.

  7. 7
    Zoom AI Companion

    Runs on isolated meeting infrastructure, dodging the chatbot API wars.

  8. 8
    Slack AI

    Stable internal orchestration, though still subject to broad cloud SaaS outages.

  9. 9
    Claude

    The most volatile major platform with 39 high-signal days in Q1 2026, buckling under massive demand surges.

  10. 10
    Microsoft 365 Copilot

    A double point of failure: when the M365 suite crashes, the AI's data source dies with it.

What Qwen noticed

Cloud providers logged 4,723 outages across 86 companies in the first half of 2026 alone, showing that AI reliability depends heavily on hyperscaler infrastructure.