Modelglass is a structured pricing and capability registry for AI models — tracked across LLM, image, video, audio, coding, science, and agentic benchmarks — built for developers and AI-native teams who need to query, compare, and route against real, versioned data instead of scattered vendor docs and stale blog posts.

Every entry starts as a YAML file: pricing, benchmark scores, context windows, modality support. Each file is schema-validated and cross-referenced against benchmark definitions, then compiled into a single content-hashed JSON artifact per vertical. That artifact is the source of truth for everything downstream — the web app, a three-tier REST API, an MCP server you can point an agent at directly, a VS Code extension for cost-aware routing, and comparison/timeline views for tracking how a model's specs shift over time.

The registries update daily via automated agents that propose changes — nothing merges without a human reviewing it first. If a price or benchmark looks stale, it's flagged, not guessed.

If you're building something that needs to pick the right model for the job — or the cheapest one that's good enough — that's what this exists for.