Anthropic's 164th outage of 2026: seven and a half hours, four models, and uptime figures that should worry enterprise buyers
A major multi-model outage on 5 August 2026 took down Claude Mythos 5, Fable 5, Opus 5, and Sonnet 5 for 7.5 hours: the 164th disruption of 2026.
At 3:05 AM ET on 5 August 2026, Claude Mythos 5, Fable 5, Opus 5, and Sonnet 5 all began returning elevated errors. By the time Anthropic’s engineers had fully resolved a second wave of degradation affecting Opus 5 specifically, the clock read 10:34 AM ET. Seven hours and 29 minutes of disruption across four of Anthropic’s flagship models, logged as the company’s 164th service disruption since January 2026.
That is not a typo. One hundred and sixty-four incidents in roughly seven months.
What actually happened on 5 August
Anthropic’s official status update was brief: “We have identified the cause of elevated errors on requests to Claude Mythos 5, Claude Fable 5, and Claude Opus 5, Claude Sonnet 5 and are working on a fix.”
The pattern underneath that statement was messier. Independent monitoring by StatusGator revealed a cascading failure rather than a clean single event. The multi-model outage began at 5:58 AM UTC. Most models showed signs of recovery by 6:28 AM UTC, but Opus 5 remained affected. By 7:33 AM UTC, Opus 5 appeared to stabilise while Sonnet 5 developed renewed errors. At 8:08 AM UTC, Sonnet 5 returned to baseline but Fable 5 began showing elevated errors. By 8:38 AM UTC, the problem had spread across all models again.
This is the hallmark of infrastructure that is under sustained pressure. Fixing one part shifts load to another, which then buckles. Anthropic’s own engineers deployed a fix and then had to open a separate incident for Opus 5 at 9:51 AM ET before finally reaching full resolution.
The numbers behind the disruption
The 5 August incident did not arrive in isolation. On 3 August, two separate events hit Sonnet 5 and multiple models. On 4 August, two more incidents occurred, including an evening disruption that took down OAuth authentication alongside model access. That event was resolved at 5:59 PM ET on 4 August, meaning enterprise developers woke up on 5 August to their second major disruption in under 12 hours.
The cumulative effect is visible in Anthropic’s own 90-day uptime figures, published on the status page as of 5 August 2026:
- claude.ai: 99.36%
- Claude API: 99.42%
- Claude Code: 99.34%
- Claude Console: 99.81%
- Claude Cowork: 99.47%
- Claude for Government: 99.99%
To put those numbers in practical terms: 99.9% uptime over 90 days allows roughly two hours of downtime per quarter per service. The claude.ai figure of 99.36% translates to approximately 13.7 hours of downtime over that same period. Claude Code at 99.34% is worse. These figures sit well below the threshold that most enterprise software contracts treat as the minimum acceptable bar.
What Anthropic’s SLA actually says
This is where it gets complicated, and where buyers need to pay close attention.
Standard API users have no contractual uptime guarantee. Anthropic’s API documentation states service is provided on a best-effort basis. Priority Tier users, meaning committed-use contracts, have a 99.5% uptime target. Enterprise customers can negotiate a 99.99% SLA, but that commitment is handled case by case rather than published as a standard contract term, and service credits are typically capped at 5 to 10 percent of monthly fees.
A 5 to 10 percent credit sounds reasonable until you calculate the actual cost of a 7.5-hour outage to a team with production workflows depending on the API. The credit rarely comes close to covering it.
For comparison, AWS, Azure, and Google Cloud all publish enforceable 99.9% uptime SLAs with financial penalties as the default for every customer, not as a negotiated exception.
The timing is awkward for Anthropic
The outage landed the same day Blackstone was reported to be pitching a second $36 billion debt deal to finance Anthropic’s procurement of Google AI chips. This follows the initial $35 billion deal Apollo Global Management and Blackstone completed roughly two months prior, structured around a special-purpose vehicle that acquires Google’s custom tensor processing units and leases them back to Anthropic.
Combined, that is approximately $71 billion in chip financing. The theory is straightforward: more compute capacity means more headroom to handle peak demand without cascading failures. Anthropic confirmed as much in an April 2026 Fortune interview: “Demand for Claude has grown at an unprecedented rate, and our infrastructure has been stretched to meet it, particularly at peak hours.”
The honest caveat is that financing a chip deal and having those chips operational are separated by months of procurement, deployment, and integration. The compute investment is real. The relief it brings to uptime figures is still on the way.
What this means for you
If you are an individual developer experimenting with the Claude API, an occasional outage is inconvenient but manageable. You adjust, retry, move on.
If your team has built production workflows around Claude, or if you are evaluating Anthropic for an enterprise contract, the picture requires more scrutiny. A few things worth acting on:
Check the status page before committing. The Claude status page publishes the 90-day uptime figures and full incident history. Read them. Not the marketing materials, the actual uptime numbers.
Ask about SLA terms in writing. If a sales team tells you Anthropic offers a 99.99% SLA, ask for the contract language, the credit structure, and the process for claiming credits. A verbal commitment is not the same as an enforceable term.
Consider running Claude via AWS Bedrock. Anthropic’s models are available through Amazon Bedrock, which applies Amazon’s own SLA structure to Claude API calls. You are still using Claude, but you gain Amazon’s reliability commitments and credit mechanisms on top.
Build for failure. If Claude is in your critical path, design your system with fallback logic. That means a secondary model provider, retry logic with exponential back-off, and graceful degradation rather than hard failures when the API is unavailable.
The $71 billion compute investment tells you Anthropic is taking the infrastructure problem seriously. The 164 incidents in seven months tell you the problem is not solved yet. For anyone making enterprise procurement decisions, that gap is the most important thing to keep in mind.