AI applications, coding tools & model APIs · Daily incidents
AI incidents: Claude Fable 5.1 and Fireworks Qwen recoveries — 8 November 2026
Anthropic reports a 14-minute Claude Fable 5.1 disruption; Fireworks resolves Qwen 3.8 Max degradation. Verified timelines and outstanding follow-ups.
Publisher disclosure: MUVO offers competing music, remix, and video tools. This report summarizes public sources and provider documentation; it is not an independent product test or a claim that MUVO is free of similar issues. Read our reporting method.
Two resolved incidents at the research cutoff
Anthropic reports that Claude Fable 5.1 recovered after a 14-minute disruption on 8 November. Fireworks separately resolved degraded performance for its hosted Qwen 3.8 Max service. These are provider-confirmed events with different scopes and timeline evidence.
This edition covers 5 November at 14:35 through 8 November at 12:12 Europe/Istanbul (UTC+3), including the 48-hour overlap from the previous successful scan. All event times below are Istanbul time. Earlier findings are deduplicated against 7 November’s report. A resolved status does not establish that every affected account or previously failed task recovered.
Claude Fable 5.1: 14-minute impact window, followed by a recovery notice
Anthropic says impact ran from 11:18 to 11:32 on 8 November, equivalent to 08:18–08:32 UTC. Its investigation named elevated errors on requests to Claude Fable 5.1; the resolution names claude.ai, the Claude API, Claude Code and Claude Cowork. This does not establish that every Claude model or request failed. Anthropic’s original incident notice.
| Milestone | 8 November, Istanbul |
|---|---|
| Provider-stated impact begins | 11:18 |
| First public acknowledgment | 11:28 |
| Provider-stated impact ends | 11:32 |
| Fix deployed; monitoring reported | 11:39 |
| Resolution notice published | 11:54 |
The 14 minutes measures Anthropic’s explicit impact boundaries, not the interval from acknowledgment to resolution. Its later monitoring and resolution notices are distinct milestones. The provider did not disclose a technical cause, regional or plan-specific breakdown, or a user workaround. MUVO has not independently measured the affected requests.
Fireworks Qwen 3.8 Max: brief recorded degradation resolved
Fireworks acknowledged impaired performance for Qwen 3.8 Max at 05:39 and marked the service operational at 05:42 on 8 November. The notice describes Fireworks’ hosted model service; it is not evidence of an outage across Alibaba’s own Qwen products. Fireworks’ original incident notice.
Its published component-impact record spans 05:39:24.835–05:42:45.507 Istanbul, a recorded interval of approximately 3 minutes 21 seconds. Those boundaries coincide with its investigation and resolution updates. They do not independently establish when every user’s symptoms began or ended.
Fireworks did not identify a particular symptom beyond degraded performance, affected plans or regions, a technical cause, or a workaround. This is separate from the earlier GLM endpoint incidents and the longer-running deployment, training and batch-inference notice.
Earlier OpenAI and Fireworks issues remain under monitoring
Neither continuing notice had a new resolution at this cutoff. OpenAI’s U.S. business Mastercard payment issue still carries the monitoring update described yesterday. Fireworks’ deployment, training and batch-inference notice still reports restored service under monitoring. See the earlier timelines and payment advice; these are follow-ups, not newly discovered outages.
Sources
The original notices are linked beside each finding. We also rechecked Google AI Studio’s incident history, Google Cloud’s history, ElevenLabs’ U.S. and EU histories, other provider archives, and accessible customer-review and community sources. No additional material Google or ElevenLabs incident was established in this scan. Incomplete archives, blocked review pages and small review samples limit coverage; absence here is not proof that every service or account worked normally. See our reporting method.
Practical analysis: Before retrying a failed paid operation, check its existing task and payment state. Recovery of a service does not prove that an earlier job was completed, canceled or refunded.
Update timeline
- 8 November 2026, 12:12 Istanbul / 09:12 UTC: Research cutoff. Verified both resolutions and rechecked continuing incidents.
- 8 November 2026: Published this edition. Product-news evidence remains queued for the separate weekly roundup.