# Anthropic Cost Monitoring: The Why, What and How? With Claude becoming central to many team workflows, Anthropic has introduced several ways to track usage and manage costs. This guide outlines the monitoring capabilities across Claude Pro, Claude API, Claude Code CLI, Claude Enterprise, and AWS Bedrock. Depending on the use case, different access methods are available: - Claude Pro – Ideal for individuals and general use cases. Offers basic plan and usage visibility. - Anthropic API – Used for integrating Claude into custom workflows or applications at scale. Includes usage tracking at the workspace level. - Claude Code – Designed for developers working with Claude through a CLI, with telemetry support and cost tracking. - Claude Enterprise – Unlocks advanced governance, audit logs, and dashboarding via custom agreements. - AWS Bedrock – Suitable for enterprises running Claude on AWS, with access to CloudWatch and detailed billing integration. Monitoring features vary by interface, but generally include workspace tracking, filters, and telemetry support through Claude Code. While it lacks some advanced features like per-user breakdowns and a usage API, Anthropic’s monitoring has evolved to cover most team-level and enterprise needs. | Feature | Claude Pro | Anthropic API | Claude Code CLI | Claude Enterprise | AWS Bedrock | | --- | --- | --- | --- | --- | --- | | Basic billing visibility | ✅ | ✅ | ✅ | ✅ | ✅ | | Workspace-level usage views | ❌ | ✅ | ✅ | ✅ | ✅ | | Filters by model or API key | ❌ | ✅ | ✅ | ✅ | ✅ | | Per-user attribution | ❌ | ❌ | ❌ | ✅ (via audit logs) | ✅ (via tags) | | Real-time metrics | ❌ | ❌ | ✅ | ❌ | ✅ | | CSV/Log exports | ❌ | ✅ | ✅ | ✅ | ✅ | | Telemetry integration | ❌ | ❌ | ✅ (OTel) | ❌ | ✅ (CloudWatch) | | Cost and usage breakdowns | Basic only | ✅ | ✅ | ✅ | ✅ (via CUR) | | Tag-based attribution | ❌ | ❌ | ❌ | ❌ | ✅ | | Alerts & anomaly detection | ❌ | ❌ | Limited (OTel) | ❌ | ✅ | | Custom dashboards | ❌ | ❌ | External only | ✅ | ✅ | ## Claude Pro (and Max) Claude Pro (and Max) are geared toward individual users. These subscriptions offer very limited monitoring features. Users can: - Check billing info and plan status - View basic usage restrictions However, there’s no detailed dashboard, session view, or telemetry available in Claude Pro/Max. ## Monitoring with the Anthropic Console (Claude API) The Console is the main dashboard if you’re using Claude via the API. It supports: - Workspace-level usage and cost views - Filters by date, model, and API key - Role-based access to usage and billing pages (updated to include Developer role access to Cost page) - CSV exports for reporting The Usage tab displays a tabular view of API activity by model, API key, and total tokens used per day. You can group usage by workspace or model, filter by time window, and download filtered views as CSV. The Cost tab provides cumulative and daily breakdowns of spending, separated by Claude model versions (e.g., Claude 4 Opus, Claude 4 Sonnet, Claude 3.5 Haiku, Claude 3.5 Sonnet). This view helps identify trends in token usage and cost anomalies. Developers can see usage charts, while Billing and Admin roles can access detailed cost breakdowns. Note that as of recent updates, the Developer role now has access to the Cost page as well (Console Role Updates). Workspaces help segment usage by team or application. Each workspace can have its own API keys, usage limits, and members. However, the Console has several limitations: - No user-level attribution - No real-time alerts or automated usage thresholds - No API-based access to monitoring data - No anomaly detection or cost forecasting features - Limited customization for visualizations or dashboards ## Claude Code: Advanced Usage Monitoring Claude Code, Anthropic’s agentic command-line tool, offers the most sophisticated monitoring capabilities in the Anthropic ecosystem through OpenTelemetry integration (https://docs.anthropic.com/en/docs/claude-code/monitoring-usage#security%2Fprivacy-considerations).Claude code gives you basic info about tokens consumed, total duration run, lines added and more information right for the terminal itself. Use the command “/cost” to get this information.  ### OpenTelemetry Integration Claude Code supports comprehensive monitoring through OpenTelemetry (OTel) metrics, providing detailed insights into usage patterns and costs. This integration allows organizations to export metrics to their preferred monitoring backends. Note: Some users have reported implementation challenges with the OpenTelemetry integration not working as expected despite proper configuration. It’s recommended to test your setup thoroughly and consider third-party monitoring tools as alternatives if needed. ### Setting Up Monitoring Install and Configure Claude Code - Follow [installation instructions](https://docs.anthropic.com/claude/docs/claude-code) - Authenticate with your Anthropic account - Ensure you have permission to configure telemetry Enable Telemetry Collection Add these variables to your shell profile (.bashrc, .zshrc, etc.): Replace the values here, with your corresponding infomation. ``` export CLAUDE_CODE_ENABLE_TELEMETRY=1 export OTEL_METRICS_EXPORTER=otlp export OTEL_EXPORTER_OTLP_PROTOCOL=grpc export OTEL_EXPORTER_OTLP_ENDPOINT=http://localhost:4317 export OTEL_METRIC_EXPORT_INTERVAL=600000 export OTEL_EXPORTER_OTLP_HEADERS="Authorization=Bearer your-token" ``` For more metrics, you can refer to Anthropic’s official docs on monitoring for Claude Code. ### Third-Party Monitoring Tools If you’re looking for additional cost monitoring capabilities, Economize offers a unified dashboard that integrates AWS CUR data, custom tagging strategies, and cloud budget alerts. It’s especially useful for teams using Claude via Bedrock and looking for cross-cloud cost visibility. ## Claude for Enterprise Enterprise plans unlock several extra capabilities: - Audit logs: Record activity by user, project, and file volume - Custom dashboards: Teams can build usage views using exported logs - Workspace enforcement: Centralized controls over access and budgets These features are not widely documented but are available via support and custom agreements. ## Claude via AWS Bedrock Using Claude through AWS Bedrock opens the door to mature AWS monitoring and billing tools that are tightly integrated into the AWS ecosystem. This setup is ideal for organizations that already rely on AWS infrastructure for operations, governance, and cost visibility. As of 2025, AWS Bedrock supports the latest Claude models including Claude 4 Opus and Claude 4 Sonnet, as well as Claude 3.5 Haiku and Claude 3.5 Sonnet with features like prompt caching now generally available. One major advantage of using Claude via Bedrock instead of the Anthropic API is the depth of observability and cost analytics available. With the Anthropic API, users are currently limited to workspace-level views, downloadable CSVs, and no native support for user-level attribution or external telemetry. Bedrock, on the other hand, integrates directly with AWS tools like CloudWatch, CUR, and tagging systems, offering automated, fine-grained, and scalable cost and usage tracking. ### CloudWatch Integration Bedrock integrates seamlessly with Amazon CloudWatch, offering comprehensive monitoring capabilities: - Real-time Metrics: Track token usage, model invocations, and activity patterns in real-time through CloudWatch dashboards. - Custom Dashboards: Create tailored monitoring views focusing on metrics most relevant to your applications and use cases. - Alerting: Set up automated alerts for cost spikes, unusual token consumption, or high session volumes. ### Setting up CloudWatch Metrics for AWS Bedrock The steps below help you configure CloudWatch to effectively monitor Claude usage on Bedrock: Enable CloudWatch Metrics - Navigate to the AWS Bedrock Console - Go to “Settings” → “Model access” - Ensure CloudWatch metrics are enabled for your models - Configure metric collection intervals (recommended: 1-minute granularity) You can go through the list of available metrics through bedrock here. Create Custom Dashboards - Open the CloudWatch Console - Navigate to “Dashboards” → “Create dashboard” - Add widgets for key Bedrock metrics: AWS/Bedrock/InvocationCount: Model invocation frequency AWS/Bedrock/InputTokenCount: Input token consumption AWS/Bedrock/OutputTokenCount: Output token generation AWS/Bedrock/InvocationLatency: Response time metrics - AWS/Bedrock/InvocationCount: Model invocation frequency - AWS/Bedrock/InputTokenCount: Input token consumption - AWS/Bedrock/OutputTokenCount: Output token generation - AWS/Bedrock/InvocationLatency: Response time metrics Configure Alarms and Alerts - In CloudWatch, go to “Alarms” → “Create Alarm” - Choose a metric such as AWS/Bedrock/InputTokenCount - Set a threshold (e.g., > 100K tokens in 5 minutes) - Choose actions (SNS notification, Lambda trigger, etc.) for alerting or auto-remediation You can also create composite alarms and link them with budgets or operational rules using AWS Budgets or EventBridge. Tagging Strategy for Governance To support chargebacks and usage attribution, apply standardized AWS resource tags to Claude Bedrock usage: - environment: (e.g., production, staging) - team: (e.g., growth, engineering) - owner: Responsible contact - project: Project or application name These tags appear in CloudWatch metrics and in Cost and Usage Reports, and help segment usage by team or initiative. Use AWS Tag Policies via AWS Organizations to enforce consistency. ### Cost and Usage Reports (CUR) CUR provides detailed billing breakdowns for every Claude usage event through Bedrock. If you’re unfamiliar with how CUR works or want to set it up from scratch, we’ve put together a detailed guide on creating AWS Cost and Usage Report. While the guide does not cover Claude specifically, it walks through enabling CUR in your AWS Billing preferences, storing reports in S3, querying usage data with Athena, and visualizing reports with QuickSight steps that are applicable for analyzing Claude spend on Bedrock. ### Tagging for Attribution Claude model calls via Bedrock support AWS-native tagging. Use these tags to categorize usage: - department - project - environment - owner Tags propagate into both CloudWatch and CUR, helping you assign cost ownership and enforce chargeback policies. Tag policies can be managed using AWS Organizations for consistency across teams. ## Best Practices and metrics to monitor: - Baseline usage: Know your typical token usage, costs, and workloads - Alerting: Use Bedrock alerts or Claude Code dashboards to catch spikes - Reviews: Weekly usage reviews and monthly cost reviews help avoid drift - Team separation: Use workspaces, tags, and keys to isolate usage - Security: Review audit logs and limit access to billing data - Test monitoring setup: Verify OpenTelemetry integrations work properly before relying on them ## What Anthropic Lacks (For Now) While Anthropic has made significant improvements, several gaps remain: - No per-API-key tracking: Unlike OpenAI, Anthropic doesn’t provide usage tracking by individual API keys, making it difficult for organizations to attribute costs to specific applications or teams - Limited real-time visibility: Anthropic’s monitoring is primarily monthly/daily rather than minute-by-minute - No programmatic API: Anthropic doesn’t offer a dedicated Usage API like OpenAI’s comprehensive tracking system - Organizational complexity: OpenAI handles sub-organizations and consolidated reporting better than Anthropic’s current workspace system However, Anthropic has been rapidly closing this gap, particularly with recent updates to their Console dashboard and the introduction of sophisticated monitoring capabilities for Claude Code. ## Final Thoughts Anthropic is making strong progress in monitoring and cost control, especially through Claude Code and AWS integrations. While OpenAI still leads in some areas like real-time APIs and user-level attribution Anthropic is still catching up to it’s competiton. If you use Claude for your business usecases , it’s worth investing in structured monitoring, either through telemetry, CloudWatch, or consolebased workspaces. Better visibility leads to better control and ultimately, more predictable outcomes for your AI budgets. --- *Source: https://www.economize.cloud/blog/anthropic-cost-monitoring*