Replicate
inference_host
62 signals tracked
Replicate: Degraded A100 Hardware — Monitoring Incident
Replicate incident (monitoring, minor impact):
Date not specified
HighIncidentReplicate: Partial System Degradation - Investigation Required
Status indicator: minor. Partial System Degradation
Date not specified
MediumIncidentReplicate: All Systems Operational
Status indicator: none. All Systems Operational
Date not specified
InfoCapabilityCog adds MCP Server Auto-Discovery Feature
Replicate introduced automatic server discovery via /.well-known/mcp/server.json endpoint, enabling easier server metadata publication and client integration
Date not specified
MediumCapabilityGoogle improves Model Reliability and Documentation with New Cog Runtime
Enhanced reliability for google/nano-banana models, improved documentation search accessibility, and added automated documentation discovery for LLMs.
Date not specified
HighCapabilityCog Predictions API adds filter for 'web' source predictions
API now supports filtering predictions by source 'web', allowing users to view only web interface generated predictions. Limited to last 14 days.
Date not specified
MediumCapabilityFlux Schnell Infrastructure Failure Partially Resolved — Monitoring Ongoing
Infrastructure issue preventing model predictions has been partially addressed. Systems are being monitored for continued performance.
Date not specified
MediumIncidentMultiple Models Prediction and Training Unavailable — Critical Incident
Inference and training systems experiencing widespread disruption. Multiple models cannot receive prediction or training requests correctly.
Date not specified
CriticalIncidentCog Playground: New Video and Image Rendering Features
Added video support for before/after slider and pixelated image rendering with size controls
Date not specified
HighCapabilityCog adds approximate cost tracking to dashboard
Added approximate cost display for predictions and trainings on dashboard pages
Date not specified
MediumCapabilityCog introduces Code Mode for Local MCP Server — experimental TypeScript code execution
Experimental feature allowing language models to write and execute TypeScript code in a sandboxed environment, with two tools: SDK documentation search and code execution
Date not specified
HighCapabilityCog introduces API Prediction Source Filtering
Ability to filter predictions API endpoint by 'web' source, showing only web interface predictions limited to last 14 days
Date not specified
HighCapabilitySystem Status: Operational
Date not specified
InfoOtherReplicate enhances model reliability and adds automatic documentation
Replicate enhanced the reliability of google/nano-banana models and improved search bar accessibility across their platform. They also added automatic llms.txt generation for documentation to help language models better understand Replicate's docs, and published guidance on running Retro Diffusion pixel art models.
Date not specified
LowCapabilityReplicate adds FLUX.2 models and improved model UI
Replicate added new FLUX.2 models (pro, flex, and dev) to their playground and made FLUX.2 pro the default model. They also enhanced model pages with created/updated dates and improved pricing displays for better clarity on bring-your-own-token models. These are additive features that don't require any action from existing users.
Date not specified
LowCapabilityReplicate updates: Cost display, homepage redesign, and Nano Banana Pro default
Replicate introduced approximate cost display for predictions and training runs on dashboard and predictions pages. The platform also launched an updated homepage with improved navigation and code snippet display, and made Nano Banana Pro the default model in the playground. These are informational updates that enhance user experience without requiring any action.
Date not specified
InfoCapabilityReplicate adds TypeScript code execution mode to localMCP
Replicate introduced an experimental code mode for their localMCP server that enables language models to write and execute TypeScript code in a sandboxed Deno environment. The feature provides SDK documentation search and code execution tools, activated using the --code-mode flag. This is an optional enhancement that doesn't affect existing functionality.
Date not specified
MediumCapabilityReplicate adds video comparison and enhanced UI controls
Replicate introduced video support for before/after comparisons, pixelated image rendering controls for pixel art models, and improved aspect ratio handling. The platform also overhauled focus states across the UI library for better accessibility. These are additive features that enhance the user experience without requiring any action from existing users.
Date not specified
LowCapabilityReplicate: Deployment failure notifications and customizable timeouts for Enterprise
Enterprise plan customers can now receive automatic email notifications when deployments fail during setup and configure custom timeout periods for model initialization. These features help catch deployment issues earlier and accommodate models requiring longer setup times for downloading large files or loading weights. Both features can be configured in deployment settings.
Date not specified
MediumCapabilityReplicate launches Python SDK public beta with HTTP API support
Replicate has released a public beta of their Python SDK, providing full support for all HTTP API operations to simplify AI model integration. The update also includes improved playground handling for different output types and streaming, plus documentation search shortcut fixes. This is a new capability enhancement that requires no immediate action from existing users.
Date not specified
InfoCapability
Get alerts for Replicate
Never miss a breaking change. SignalBreak monitors Replicate and dozens of other AI providers in real time.
Sign up free — no credit card required