LLM Providers Status Report - December 2025

Summary

This report covers the operational status of major AI systems during December 2025, including Anthropic, Cohere, DeepSeek, Google Gemini, Groq Cloud, OpenAI, Perplexity, Replicate, and xAI. The data encompasses official incidents reported by providers and monitoring insights from IsDown's detection system.

Important: Each provider has their own framework for reporting incidents. A provider showing more incidents does not mean they are less reliable - it may indicate more transparent reporting. These providers also vary significantly in size and market share, making direct comparisons inappropriate.

December 2025 Status Overview

Detailed Provider Analysis

Anthropic

Anthropic reported 20 incidents in December 2025 (7 major, 13 minor) with a total incident time of approximately 184.5 hours and an average resolution time of 553 minutes per incident.

The incidents primarily consisted of elevated error rates across various Claude models (Haiku 4.5, Opus 4.5, Sonnet 4, and Sonnet 4.5), with multiple occurrences throughout the month. Several service availability issues affected claude.ai, including complete unavailability on December 5th and inability to create new conversations on December 16th. iOS subscription access problems persisted for an extended period, with a bug preventing proper plan activation for Apple App Store customers who purchased after November 24th. Additional issues included documentation site errors in Europe, shared chat viewing problems, and Windows rendering issues for Claude Code.

Cohere

Cohere reported no incidents for December 2025.

DeepSeek

DeepSeek reported 4 incidents in December 2025 (3 major, 1 minor) with a total incident time of approximately 2.0 hours and an average resolution time of 29 minutes per incident.

The incidents were characterized by web and API service unavailability, with three major outages occurring on December 11th, 14th, and 16th. Each of these lasted approximately 30-35 minutes. A minor performance degradation incident affecting web and API services occurred on December 22nd and was resolved in about 14 minutes.

Google Gemini

Google Gemini reported no incidents for December 2025.

Groq Cloud

Groq Cloud reported no incidents for December 2025.

OpenAI

OpenAI reported 22 incidents in December 2025 (1 major, 21 minor) with a total incident time of approximately 182.7 hours and an average resolution time of 498 minutes per incident.

The incidents encompassed a wide range of service disruptions. ChatGPT experienced multiple error rate increases affecting various platforms including web, Android devices, and business/enterprise accounts. Authentication and connector issues were prominent, with GitHub authentication problems for Codex Web persisting for extended periods and ChatGPT connectors unexpectedly disconnecting, requiring several days to fully restore. Sora services faced login issues and API errors. Additional problems included fine-tuning job failures, image API errors, SSO authentication difficulties, and conversation history loading issues. Several incidents involved specific model versions, particularly affecting 5.2 Auto and 5.2 Instant models.

Perplexity

Perplexity reported 1 incident in December 2025 (1 major, 0 minor) with a total incident time of approximately 0.6 hours and an average resolution time of 35 minutes per incident.

The single incident was a service outage that occurred on December 5th, lasting approximately 36 minutes.

Replicate

Replicate reported no incidents for December 2025.

xAI

xAI reported no incidents for December 2025.

Conclusion

During December 2025, 47 total incidents were reported across all monitored AI systems, with Anthropic and OpenAI accounting for the majority of reported incidents. Five providers (Cohere, Google Gemini, Groq Cloud, Replicate, and xAI) reported zero incidents throughout the month. The most common resolution time across all incidents was 30-60 minutes, though OpenAI's incidents typically required 4+ hours to resolve. Friday and Tuesday were the days with the highest incident counts across providers.