July 29–30, 2026: @ClaudeDevs confirmed two separate incidents in a 24-hour window that cut into Anthropic's capacity to serve Claude — elevated errors and reduced availability across claude.ai, the API, Claude Code, and Claude Cowork.
"Two incidents over the last 24 hours resulted in elevated errors or reduced availability on Claude for some users. Multiple separate network failures cut into our capacity to serve Claude, and some requests failed while we rerouted traffic."
Services were fully recovered by late July 30, 2026. Claude for Government stayed online throughout — it runs on isolated infrastructure separate from the consumer and developer surfaces that took the hit.
TL;DR — outage at a glance
| Detail | Value |
|---|---|
| Window | July 29–30, 2026 (~24 hours, two incidents) |
| Cause | Multiple separate network failures cut serving capacity |
| Symptom | Elevated errors / reduced availability, some failed requests during reroute |
| Surfaces hit | claude.ai, API, Claude Code, Claude Cowork |
| Unaffected | Claude for Government — online throughout |
| Recovery | Full by late July 30, 2026 |
| Source | @ClaudeDevs status update |
| Comparable prior incident | July 17 Fable access glitch — ~30 min, different root cause |
Timeline
| When | Event |
|---|---|
| July 29, 2026 | First incident begins — network failure cuts serving capacity; elevated errors reported |
| ~30 min into one incident | @theo (t3.gg): "Claude Code has been mostly down for half an hour. Really inconvenient, but I'll forgive them if we get a reset" |
| July 29–30 | Second, separate network failure compounds the disruption; requests fail during traffic reroute |
| Late July 30, 2026 | Full recovery confirmed across claude.ai, API, Claude Code, and Claude Cowork |
| Ongoing | Claude for Government unaffected the entire window |
Two distinct network failures — not one prolonged event — is why some users saw the outage as intermittent rather than continuous. Anthropic's own framing ("two incidents over the last 24 hours") matches that pattern: capacity dipped, partially recovered, then dipped again before the final fix held.
What actually broke
Anthropic's statement is specific about mechanism, not root cause:
- Network failures — plural, separate events — reduced the pool of infrastructure available to serve requests.
- Rerouting traffic around the failed paths itself caused some requests to fail, a common side effect of live capacity migration under load.
- The impact was capacity-shaped: "elevated errors or reduced availability" reads as partial degradation for many users rather than a hard, total outage for everyone simultaneously — consistent with Theo's "mostly down" description rather than a full blackout.
No public postmortem with root-cause detail (which network layer, which region, which provider) had been published as of this writing. That is typical for a same-week status update; a fuller writeup, if one comes, usually follows days later.
X reaction — grass, resets, and déjà vu
The outage produced the now-familiar split between practical frustration and gallows humor:
| Reaction | Who | Tone |
|---|---|---|
| Grim tolerance | @theo | "I'll forgive them if we get a reset" |
| Touch grass jokes | CJ The X, ℏεsam | Screenshots of a Polymarket market tied to the outage, captioned "run into the woods" |
| Industry-wide shrug | Multiple | "The entire tech industry when Claude has an outage" — memes about how much of the dev ecosystem now runs through Claude Code |
The joke cuts both ways: an outage funny enough to meme is also evidence of how load-bearing Claude Code has become for daily engineering work. Compare that dependency to the July 17 Fable access glitch — a 30-minute billing bug that produced a nearly identical reaction cycle two weeks earlier.
Is this part of a pattern?
2026 has been a rough year for trusting the dashboard, not just for Anthropic:
| Date | Incident | Root cause | Duration |
|---|---|---|---|
| Jul 17, 2026 | Fable 5 access glitch | Billing/access gate bug | ~30 min |
| Jul 29–30, 2026 | This outage | Multiple network failures | ~24 hrs (two incidents) |
| Same period | AWS Cost Explorer billing bug | Display/estimation error | Hours |
Different root causes, same underlying story: demand on frontier AI infrastructure is outrunning the margin for error in billing systems and network capacity alike. None of these incidents individually is catastrophic; the frequency is what's notable.
Why frontier-model capacity incidents keep happening
Anthropic doesn't publish raw request-volume figures, but the shape of 2026's incident log is consistent with a system running close to its ceiling rather than one with slack to absorb a bad network day. A few structural reasons this keeps recurring across the industry, not just at Anthropic:
- Demand growth is front-loaded, capacity growth is not. Model releases (Opus 5, GPT-5.6, Gemini 3.6) each pull in a wave of new agentic workloads — Claude Code sessions, Codex runs, autonomous pipelines — that show up as request volume the day a model ships. Data center capacity, by contrast, is provisioned months ahead on forecasts that are frequently wrong in the optimistic direction.
- Agentic usage is bursty in a way chat usage wasn't. A single Claude Code session or
/looprun can fire dozens of tool calls and follow-up completions in seconds. That concentrates load in ways a traditional chat-request pattern didn't, making capacity planning harder to get right. - Network failures compound under load. A network path failure that would be a minor blip at 60% utilization becomes a capacity event at 90%+ utilization, because there's no spare headroom to reroute into without some requests failing during the transition — exactly the mechanism @ClaudeDevs described.
- Billing and access systems are newer than the core serving stack. The July 17 Fable glitch was a credit-gating bug, not a network failure — a different subsystem, but the same underlying story of infrastructure built and iterated on faster than it can be fully hardened.
None of this excuses a 24-hour disruption window, but it explains why "elevated errors" incidents have become a recurring 2026 pattern across frontier labs rather than a one-off Anthropic problem. Watch for similar disclosures from OpenAI and Google as Codex and Gemini agentic usage scales the same way Claude Code's has over the past several months, pulling in workloads none of these platforms were originally provisioned to absorb at this density or this speed.
How this compares to competitor uptime in July 2026
Outages are hard to compare cleanly across providers because disclosure practices differ — some labs post detailed incident writeups, others stay quiet unless pressed. What's visible from public reporting this month:
| Provider | Disclosed incident | Public detail |
|---|---|---|
| Anthropic (Claude) | Two network failures, Jul 29-30 | @ClaudeDevs statement, no root-cause postmortem yet |
| Anthropic (Claude) | Access/billing glitch, Jul 17 | Full timeline, refund policy disclosed |
| OpenAI | No comparable July disruption reported at this scale | — |
The comparison is not meant to score Anthropic against OpenAI on reliability — a single month of visible incidents is too small a sample. It's a reminder that whichever frontier lab you build on, budget for occasional capacity incidents as a cost of using cutting-edge, rapidly-scaling infrastructure rather than a mature, over-provisioned utility.
What to do if you hit errors
- Check before escalating — a 24-hour incident window means intermittent errors during that period were very likely the outage, not your setup.
- Restart Claude Code if a session is stuck mid-task; reselect your model with
/modelif it reset to a fallback. - Save context and prompts before switching harnesses — if you fell back to GPT-5.6 Sol in Codex mid-outage, don't duplicate the same work across both once Claude recovers.
- Don't assume billing impact — this was a capacity/network incident, not the credit-gate bug from July 17; there's no indication of erroneous overcharges tied to this outage.
- Watch @ClaudeDevs directly for the fastest confirmation during any future incident — it was first to confirm both the fault and the recovery here.
Summary
Two separate network failures over July 29–30, 2026 cut into Anthropic's capacity to serve Claude, producing elevated errors and reduced availability on claude.ai, the API, Claude Code, and Claude Cowork. Claude for Government was unaffected. Anthropic confirmed full recovery by late July 30. No root-cause postmortem has been published yet. It's the second notable Claude availability incident in two weeks, following the July 17 Fable access glitch — different causes, same reminder that frontier-model infrastructure is being run close to the edge of its capacity.
Related on explainx.ai
- Fable 5 access glitch — July 17, 2026 outage explained
- Claude usage limits — 2026 timeline explained
- Claude, Codex, Cursor usage limits reset — July 2026
- AWS billing glitch — trillion-dollar Cost Explorer bug
- Claude Code desktop + browser built in
- Anthropic IPO S-1 and investor meetings — July 2026
- GPT-5.6 Sol + Claude Code (Claudex) setup guide
- OpenAI slashes GPT-5.6 prices up to 80% — July 2026
Official source: @ClaudeDevs status update
Incident details reflect @ClaudeDevs' public statement and social reporting as of July 31, 2026. Check current Claude status directly before treating any residual errors as a continuation of this incident.