SynthesisFrontier AI and Model Releases1h ago2 sources2 min readPrimary: Claude Status
Published Mar 25, 2026, 3:22 PM UTC
TLDR
Treat this as a claude.ai web UI transport instability centered on Cowork: keep API usage unchanged, add retry/backoff for UI-driven calls, and monitor the two active incidents for scope changes and resolution updates.
Topic context
Use this page when you want fast context on confirmed model launches from OpenAI, Anthropic, Google DeepMind, xAI, Meta, and similar labs without scanning every release note, model card, or developer post yourself. Key angles: openai, anthropic, google deepmind, gemini.
openaianthropicgoogle deepmindgeminiclaudegpt
Anthropic posted two active incidents—general elevated errors on claude.ai and elevated connection reset errors in Cowork—both under investigation with minor/none stated impact, which together suggest a transport/UI-layer issue rather than a model outage or API-wide regression.
What Changed
- Anthropic reported “Elevated Errors on claude.ai” with status Investigating and impact listed as None [2].
- A separate, concurrent incident cites “Elevated connection reset errors in Cowork,” also Investigating with impact labeled Minor [1].
- Both updates are recent and limited to claude.ai/Cowork; no mention of model degradation or API-wide outage in either note [1][2].
Cross-Source Inference
- Scope likely UI-side: The pairing of a broad claude.ai errors notice with a more specific Cowork connection reset description implies the web UI/experience is principally affected, not underlying model inference or auth flows (medium confidence) [1][2].
- Transport-layer pattern: The explicit “connection reset” language points to network/edge or session termination behavior rather than model runtime crashes; absence of token/output anomalies supports this (medium confidence) [1][2].
- Limited customer impact so far: Impact fields marked None/Minor suggest a constrained blast radius and intermittent failures, not sustained downtime (high confidence) [1][2].
- API unaffected for now: With no API incident posted alongside these UI notices, integrators can assume normal API operations pending new signals, but should monitor for changes (medium confidence) [1][2].
Implications and What to Watch
- Short-term operations: Maintain normal API traffic; for UI-driven or browser-proxied calls, enable exponential backoff and circuit breakers to ride out resets.
- Monitoring: Track both incident pages for resolution notes or scope expansion; watch for any new API-status items that would change risk posture [1][2].
- Evidence gaps: No geographic segmentation, root cause, or timeline provided; await post-incident details to confirm whether this was CDN/edge, session management, or another UI-layer dependency.