Amazon Bedrock now supports observability of First Token Latency and Quota Consumption

This article announces two new CloudWatch metrics for Amazon Bedrock to improve observability of AI inference performance and resource usage.

• TimeToFirstToken measures streaming API latency from request to first token receipt
• EstimatedTPMQuotaUsage tracks Tokens Per Minute quota consumption across all inference APIs
• Both metrics enable CloudWatch alarms for performance monitoring and quota management
• Available in all commercial Bedrock regions with no client-side instrumentation required
• Metrics update every minute for successfully completed requests at no additional cost

These new metrics provide deeper visibility into Bedrock inference performance and quota consumption without requiring code changes or opt-in configuration.