Changelog
Unified Billing now supports opt-in Zero Data Retention. This ensures supported upstream AI providers (eg OpenAI ZDR) do not retain request and response data.
- Supports adding OpenAI compatible Custom Providers for inferencing with AI providers that are not natively supported by AI Gateway
- Cost and usage tracking for voice models
- You can now use Workers AI via AI Gateway with no additional configuration. Previously, this required generating / passing additional Workers AI tokens.
Unified Billing
- Unified Billing is now in open beta. Connect multiple AI providers (e.g. OpenAI, Anthropic) without any additional setup and pay through a single Cloudflare invoice. To use it, purchase credits in the Cloudflare Dashboard and spend them across providers via AI Gateway.
- Add support for pipecat model on Workers AI
- Fix OpenAI realtime websocket authentication.
- Added cost tracking and observability support for async video generation requests for OpenAI Sora 2 and Google AI Studio Veo 3.
cf-aig-eventIdandcf-aig-log-idheaders are now returned on all requests including failed requests
The Model playground is now available in the AI Gateway Cloudflare Dashboard, allowing you to request and compare model behaviour across all models supported by AI Gateway.
- Add support for Deepgram on Workers AI using Websocket transport.
- Added Parallel as a provider.
OTEL Tracing
Added OpenTelemetry (OTEL) tracing export for better observability and debugging of AI Gateway requests.
- Added support for Fal AI provider.
- You can now set up custom Stripe usage reporting, and report usage and costs for your users directly to Stripe from AI Gateway.
- Fixed incorrectly geoblocked requests for certain regions.
- New API endpoint (
/compat/v1/models) for listing available models along with their costs. - Unified API now supports Google Vertex AI providers and all their models.
- BYOK support for requests using WebSocket transport.
Data Loss Prevention
Data loss prevention capabilities are now available to scan both incoming prompts and outgoing AI responses for sensitive information, ensuring your AI applications maintain security and compliance standards.
Dynamic routing
Introduced Dynamic routing that lets you visually or via JSON define flexible request flows that segment users, enforce quotas, and choose models with fallbacks—without changing application code.
Bring your own keys (BYOK)
Introduced Bring your own keys (BYOK) allowing you to save your AI provider keys securely with Cloudflare Secret Store and manage them through the Cloudflare dashboard.
New GA providers
We have moved the following providers out of beta and into GA:
OpenAI Compatibility
- Introduced a new OpenAI-compatible chat completions endpoint to simplify switching between different AI providers without major code modifications.
- Increased Max Number of Gateways per account: Raised the maximum number of gateways per account from 10 to 20 for paid users. This gives you greater flexibility in managing your applications as you build and scale.
- Streaming WebSocket Bug Fix: Resolved an issue affecting streaming responses over WebSockets. This fix ensures more reliable and consistent streaming behavior across all supported AI providers.
- Increased Timeout Limits: Extended the default timeout for AI Gateway requests beyond the previous 100-second limit. This enhancement improves support for long-running requests.
Cache Key Calculation Changes
- We have updated how cache keys are calculated. As a result, new cache entries will be created, and you may experience more cache misses than usual during this transition. Please monitor your traffic and performance, and let us know if you encounter any issues.
WebSockets
- Added WebSockets API to provide a persistent connection for AI interactions, eliminating repeated handshakes and reducing latency.
Guardrails
- Added Guardrails help deploy AI applications safely by intercepting and evaluating both user prompts and model responses for harmful content.
Updated Log Storage Settings
- Introduced customizable log storage settings, enabling users to:
- Define the maximum number of logs stored per gateway.
- Choose how logs are handled when the storage limit is reached:
- On - Automatically delete the oldest logs to ensure new logs are always saved.
- Off - Stop saving new logs when the storage limit is reached.
Added request handling
- Added request handling options to help manage AI provider interactions effectively, ensuring your applications remain responsive and reliable.
New AI Gateway providers
- Configuration: Added ElevenLabs, Cartesia, and Cerebras as new providers.
DeepSeek
- Configuration: Added DeepSeek as a new provider.
AI Gateway Dashboard
- Updated dashboard to view performance, costs, and stats across all gateways.
Bug Fixes
- Bug Fixes: Fixed Anthropic errors being cached.
- Bug Fixes: Fixed
env.AI.run()requests using authenticated gateways returning authentication error.
OpenRouter
- Configuration: Added OpenRouter as a new provider.
WebSockets API
- Configuration: Added WebSockets API which provides a single persistent connection, enabling continuous communication.
Authentication
- Configuration: Added Authentication which adds security by requiring a valid authorization token for each request.
Grok
- Providers: Added Grok as a new provider.
Vercel SDK
Added Vercel AI SDK. The SDK supports many different AI providers, tools for streaming completions, and more.
Persistent logs
- Logs: AI Gateway now has logs that persist, giving you the flexibility to store them for your preferred duration.
Logpush
- Logs: Securely export logs to an external storage location using Logpush.
Pricing
- Pricing: Added pricing for storing logs persistently.
Evaluations
- Configurations: Use AI Gateway’s Evaluations to make informed decisions on how to optimize your AI application.
Custom costs
- Configuration: AI Gateway now allows you to set custom costs at the request level custom costs to requests, accurately reflect your unique pricing, overriding the default or public model costs.
Mistral AI
- Providers: Added Mistral AI as a new provider.
Google AI Studio
- Providers: Added Google AI Studio as a new provider.
Custom metadata
AI Gateway now supports adding custom metadata to requests, improving tracking and analysis of incoming requests.
Logs
Logs are now available for the last 24 hours.
Custom cache key headers
AI Gateway now supports custom cache key headers.
Access an AI Gateway through a Worker
Workers AI now natively supports AI Gateway.
- Added new endpoints to the REST API.
- LLM Side Channel vulnerability fixed
- Providers: Added Anthropic, Google Vertex, Perplexity as providers.
- Real-time Logs: Logs are now real-time, showing logs for the last hour. If you have a need for persistent logs, please let the team know on Discord. We are building out a persistent logs feature for those who want to store their logs for longer.
- Providers: Azure OpenAI is now supported as a provider!
- Docs: Added Azure OpenAI example.
- Bug Fixes: Errors with costs and tokens should be fixed.
- Logs: Logs will now be limited to the last 24h. If you have a use case that requires more logging, please reach out to the team on Discord.
- Dashboard: Logs now refresh automatically.
- Docs: Fixed Workers AI example in docs and dash.
- Caching: Embedding requests are now cacheable. Rate limit will not apply for cached requests.
- Bug Fixes: Identical requests to different providers are not wrongly served from cache anymore. Streaming now works as expected, including for the Universal endpoint.
- Known Issues: There's currently a bug with costs that we are investigating.
Was this helpful?
- Resources
- API
- New to Cloudflare?
- Directory
- Sponsorships
- Open Source
- Support
- Help Center
- System Status
- Compliance
- GDPR
- Company
- cloudflare.com
- Our team
- Careers
- © 2025 Cloudflare, Inc.
- Privacy Policy
- Terms of Use
- Report Security Issues
- Trademark
-
