Skip to main content
Press releasePublished

Vibgrate benchmark measures 26.3% fewer tokens on a 300-file repository

In Vibgrate's latest pinned A/B benchmark, an AI coding assistant using Vibgrate AI Context consumed 493,895 tokens, against 670,169 with generic file discovery — 26.3% fewer across the approximately 300-file test tier.

In Vibgrate's latest pinned A/B benchmark, an AI coding assistant using Vibgrate AI Context consumed 493,895 tokens, compared with 670,169 tokens using generic file discovery — 26.3% fewer across the approximately 300-file test tier. Both runs used the same model and tasks, and only tasks successfully completed by both configurations were included.

The published results also show that the code map increased token use on very small repositories and produced larger savings on the approximately 1,000-file test case. Repository size changes the economics: mapping a five-file project costs more than it saves, while on larger codebases the map reduces repeated repository searching.

The complete figures are measured in detail on the token savings benchmark page, which publishes the methodology, task-level results, model, CLI version, and verifier approach for every pinned run.

Facts and primary sources

  • The pinned CLI 2026.819.3 benchmark records 493,895 total tokens with Vibgrate AI Context against 670,169 with generic file discovery on the ~300-file tier (26.3% fewer), counting only tasks completed successfully by both configurations, with task-level results published. Read the full token-savings benchmark

About Vibgrate

Vibgrate is a software intelligence platform that helps teams find aging dependencies, assess security exposure, and plan modernization. Its free CLI maps codebases and produces DriftScore, a transparent measure of dependency, framework, and runtime drift. Vibgrate Cloud adds portfolio visibility, governance workflows, and verifiable evidence for security and regulatory decisions. Vibgrate Evidence preserves version-specific component and release records so teams can reproduce historical exposure findings. Vibgrate AI Context gives coding agents version-matched documentation and precise repository context while reducing unnecessary token use. Vibgrate publishes its scoring methodology openly, allowing researchers to inspect the evidence, assumptions, and formulas behind its results.

25-word boilerplate

Vibgrate helps software teams find aging dependencies, assess security exposure, and plan modernization using verifiable evidence produced by its free code-scanning CLI and cloud platform.

50-word boilerplate

Vibgrate is a code drift intelligence platform for software teams. Its free CLI maps a codebase and identifies drift across dependencies, frameworks, and runtimes. Vibgrate Cloud turns the same scan data into portfolio visibility, governance workflows, security prioritization, and verifiable evidence for modernization decisions, regulatory reporting, and ongoing software assurance.

Press inquiries

Peter Chapman, Founder · press@vibgrate.com