Why 86% of Claude Code Bill Has Nothing to Do With Your Prompts
AI assets enter codebases through diverse mechanisms such as external APIs, configuration files, and loader scripts, making them difficult to track with traditional lockfiles. Checkmarx AI Inventory, a component of Checkmarx One, addresses this by scanning for five distinct AI asset types in a single pass. The tool provides detailed metadata including provider and version information, which can be exported as an AI-BOM for compliance audits. This capability has contributed to Gartner naming Checkmarx a leader in software supply chain security. Claude Code users often face unexpectedly high bills because user prompts account for only 14% of input tokens. The majority of costs stem from hidden context, including static overhead, MCP server schemas, tool results, and replayed conversation history. To address this lack of visibility, Comet introduced Opik Cost Intelligence, a tool that provides granular token attribution and automated optimization recommendations. By managing configurations like model choice and thinking effort, teams can significantly reduce expenses without sacrificing performance.
閱讀原文 ↗目錄
Catalog every AI asset type in one scan
AI assets enter codebases through diverse mechanisms such as external APIs, configuration files, and loader scripts, making them difficult to track with traditional lockfiles. Checkmarx AI Inventory, a component of Checkmarx One, addresses this by scanning for five distinct AI asset types in a single pass. The tool provides detailed metadata including provider and version information, which can be exported as an AI-BOM for compliance audits. This capability has contributed to Gartner naming Checkmarx a leader in software supply chain security.
- AI assets exist in codebases as five distinct types: LLM APIs, MCP servers, agents, tool plugins, and datasets.
- Traditional lockfiles are insufficient for tracking AI assets because they are often embedded in application code or config files.
- Checkmarx AI Inventory scans for all five AI asset types simultaneously to provide a unified security view.
- The tool generates an AI-BOM (AI Bill of Materials) that includes provider, version, and exact file locations for auditing.
- Gartner recognized Checkmarx as a Leader in its inaugural Magic Quadrant for Software Supply Chain Security.
Why 86% of Claude Code bill has nothing to do with your prompts
Claude Code users often face unexpectedly high bills because user prompts account for only 14% of input tokens. The majority of costs stem from hidden context, including static overhead, MCP server schemas, tool results, and replayed conversation history. To address this lack of visibility, Comet introduced Opik Cost Intelligence, a tool that provides granular token attribution and automated optimization recommendations. By managing configurations like model choice and thinking effort, teams can significantly reduce expenses without sacrificing performance.
- User prompts represent only 14% of input tokens in a typical Claude Code session.
- Prior assistant context, specifically replayed tool results, accounts for 30-45% of total input spend.
- MCP server schemas can add up to 16,000 tokens per turn if many servers are connected.
- Setting thinking effort to 'medium' instead of 'high' reduces output tokens by 76% with similar task completion rates.
- Sonnet models are approximately 40% cheaper than Opus while maintaining comparable performance for coding tasks.
- Comet's Opik Cost Intelligence uses a proxy plugin to track costs without storing sensitive prompt or code data.