If you were trying to use Claude yesterday and found yourself staring at a spinning loader, a login error, or a blank screen, you were far from alone.
If you were trying to use Claude yesterday and found yourself staring at a spinning loader, a login error, or a blank screen, you were far from alone. On April 28, 2026, Anthropic’s Claude AI platform suffered one of its most significant and widely reported outages of the year, knocking out Claude. ai, the Anthropic API, Claude Code, Claude Haiku 4.5, and even Claude Design simultaneously.
Well, then here is the complete, verified breakdown of what happened, when, how it was handled, and how it stacks up against ongoing reliability issues at ChatGPT and Google Gemini. Importantly, there are several details about this incident that most coverage missed entirely.
Most reports treated April 28 as a single outage event. In reality, Anthropic’s official status page at status.claude.com reveals three distinct disruptions occurred on the same day:
Incident 1 — Early morning (brief): An elevated error window ran from 6:22 PT / 13:22 UTC to 6:39 PT / 13:39 UTC — a roughly 17-minute disruption that was resolved and formally closed at 13:50 UTC. Most users in the US were asleep or not yet working. This went largely unnoticed in the press.
Incident 2 — Mid-morning (Claude Haiku 4.5 specifically): Elevated errors on Claude Haiku 4.5 were logged between 4:53 PT / 11:53 UTC and 5:36 PT / 12:44 UTC, with a fix deployed and confirmed at 12:45 UTC. This affected developers and API users relying on Haiku, Anthropic’s faster, cheaper model variant.
Incident 3 — Afternoon (the major outage): The most severe disruption began at 17:41 UTC when Anthropic posted: “We are investigating an issue preventing users from reaching Claude.ai.” Minutes later at 17:51 UTC, the scope became clear, elevated errors across the Anthropic API, access failures to Claude.ai, and broken login paths for Claude Code. This was officially labeled a “Major Outage.”
Downdetector, the crowd-sourced outage tracking platform widely used to answer the question “is Claude down?”, told the story in numbers:
The official Claude status page identified two core failure categories:
Beyond those official categories, users on X (formerly Twitter) and Reddit reported a third area of failure: Claude Design, Anthropic’s visual AI tool, was also inaccessible during the peak outage window.
Anthropic moved relatively quickly once the issue was identified. A fix was announced at approximately 2:20 PM ET, and Downdetector reports fell sharply in the minutes that followed, dropping from nearly 8,000 to a few hundred within the hour. By the time most users checked again mid-afternoon, Claude’s status page had returned to green across all components.
The total window for the major afternoon incident was roughly 40–60 minutes of widespread inaccessibility, with a tail of residual errors extending slightly longer for some users.
April 28 was not an isolated event. The past 30 days have included:
Over the past 90 days, according to IsDown — which monitors official status pages rather than just relying on crowd reports — Claude has logged 137 incidents: 43 major outages and 94 minor incidents, with a median resolution time of 1 hour and 8 minutes. Since October 2025, IsDown has tracked 197 Claude incidents in total. StatusGator, a separate monitoring service, has recorded over 466 outages affecting Claude users over the past 7 months.
Here is where most coverage of the April 28 outage falls short: Claude was not the only AI platform struggling that day, or that week.
ChatGPT experienced its own major disruption on April 20, 2026, the same date Claude had its “Service is temporarily busy” wave. At peak, roughly 5,000 users reported issues on Downdetector. The failure affected ChatGPT, Codex, and the API Platform simultaneously, with users unable to access projects and some losing access to ongoing conversation threads.
OpenAI was relatively fast to acknowledge the problem on its status page, but slower to fix it — reports on Downdetector stayed elevated for over an hour after the initial spike. A subsequent degradation event on April 22 affected enterprise and business workspaces for over 8 hours, and free users saw conversation errors for nearly 2.5 hours separately.
Over the past 90 days, OpenAI has logged 61 incidents (1 major, 60 minor) according to IsDown, with a median resolution time of 2 hours, significantly longer than Claude’s 1 hour and 8 minutes. When ChatGPT goes down, users typically wait nearly 5 hours (299 minutes) before full resolution. OpenAI’s own self-reported uptime figures range from 99.80% to 99.99%, depending on the component.
Google Gemini’s official incident log shows just 4 reported incidents in the past 90 days — a figure that, on the surface, looks dramatically better than both Claude and ChatGPT. The most recent officially acknowledged incident was in February 2026, when chat history was invisible across the Gemini web interface and mobile apps.
But here is what most coverage misses: on April 28, 2026 — the same day Claude went down — StatusGator flagged a nearly 12-hour degradation warning for the Google AI Studio and Gemini API. This was not prominently disclosed on Google’s status page and received essentially no media coverage. It is a pattern consistent with Google’s historically conservative approach to incident disclosure; the company tends to log far fewer incidents than its platforms actually experience, particularly at the API level.
A separate multi-day Gemini API incident ran from April 13 to April 17, lasting nearly 4 days and affecting developers building on the Gemini API. This also received minimal coverage.
When Gemini’s incidents are acknowledged, they tend to last longer than either Claude or ChatGPT — the median duration of officially logged incidents is 2 hours and 28 minutes.
| Claude | ChatGPT | Gemini | |
| Incidents (90 days, official) | 137 (43 major) | 61 (1 major) | 4 |
| Median resolution time | 1h 8m | 2h 0m | 2h 28m |
| Typical max resolution | ~3.25 hours | ~5 hours | Not disclosed |
| Self-reported uptime | Not disclosed | 99.80–99.99% | Not disclosed |
| Notable April incident | Apr 28, ~8,000 reports | Apr 20, ~5,000 reports | Apr 28, 12h API warning |
| Transparency of reporting | High | Medium-High | Low |
The headline figure, Claude has 137 incidents vs. Gemini’s 4, is the kind of statistic that can mislead without context.
Claude’s high count reflects transparency, not necessarily worse infrastructure. Anthropic publishes incidents granularly and quickly. A 17-minute spike in error rates at 6 AM Pacific becomes a formally logged and resolved incident. Google, by contrast, appears to have experienced a 12-hour API degradation on April 28 with no official acknowledgment whatsoever. The “4 incidents” figure for Gemini reflects what Google chooses to publish — not what actually happened.
Claude resolves problems faster than either competitor. A median of 1 hour 8 minutes is meaningfully better than ChatGPT’s 2-hour median and Gemini’s 2 hour 28 minute average. If you do hit an outage, Claude historically gets you back online faster.
The shared pattern points to an industry-wide infrastructure problem. ChatGPT, Claude, and Gemini all experienced significant disruptions within the same 10-day window in April 2026. This is not a coincidence. The explosive growth of AI coding assistants (Claude Code, Codex) and agentic tools has placed enormous and rapidly scaling demand on shared authentication infrastructure, GPU clusters, and API gateways across the industry. None of the major providers have fully built out the redundancy their current user volumes require.
Claude Code’s growth is a likely contributor. Anthropic has acknowledged that Claude Code has “exploded in popularity” in 2026. Multiple incidents — including the April 28 major outage — specifically called out “login paths for Claude Code” as a failure point, suggesting that auth and session management systems are under particular strain as the developer base scales.
The pattern across April suggests several areas where Anthropic’s reliability needs improvement:
1. Shared authentication infrastructure is a recurring weak point. The April 15 and April 28 outages both specifically broke login paths for Claude Code and Claude.ai simultaneously. This suggests a common auth dependency that is not sufficiently redundant. Anthropic should prioritize separating API authentication from consumer login flows so a failure in one doesn’t cascade into the other.
2. Model-specific failures need better isolation. The Haiku 4.5 errors on April 28 should not have been possible to notice on the consumer Claude.ai interface at all — model-level errors should be handled with automatic failover to alternate capacity, not surfaced as user-facing failures.
3. Proactive communication before Downdetector catches it. In the April 28 afternoon incident, users were filing reports on Downdetector 10–15 minutes before the official status page was updated. That gap erodes trust unnecessarily. Anthropic should invest in faster detection and communication tooling.
4. Enterprise-grade SLAs need infrastructure to match. As Anthropic pushes deeper into enterprise sales, the current incident frequency — roughly 1.5 incidents per day on average — is a significant friction point. Enterprise customers building on the API need contractual uptime commitments backed by actual redundancy.
If you are wondering “is Claude down?” in real time, here are the best sources to check:
If Claude.ai is down and you need to keep working, Claude is also accessible via the iOS and Android apps, which sometimes remain functional when the web interface is degraded. The Claude API and claude.ai consumer interface share infrastructure in many areas, but not all.
Claude went down on April 28, 2026, three times, in fact, though only the afternoon incident made headlines. At peak, nearly 8,000 users reported access failures across Claude.ai, the API, Claude Code, and Claude Design. The fix arrived roughly an hour after the main disruption began, and services were restored by mid-afternoon ET.
The outage is part of a broader pattern: Claude has logged 137 incidents in the past 90 days, the most of any major AI platform, but also resolves them fastest, at a median of 68 minutes. ChatGPT logged 61 incidents in the same window and takes twice as long to recover on average. Gemini officially reports just 4 incidents, but its actual degradation rate appears significantly higher based on third-party monitoring data.
The deeper story is that the entire AI industry is running its infrastructure at the edge of its capacity, as user adoption, particularly for developer tools like Claude Code and Codex, accelerates faster than redundancy can be built. Until that changes, the answer to “is Claude down?” will periodically, and frustratingly, be yes.
For live Claude status, visit: status.claude.com.
For historical incident comparisons, see: isdown.app
