What these metrics mean and why they matter
The inflation rate for AI work. Tracks the cost of a standardized basket of AI workloads against a January 2025 baseline. A value of 87 means AI work is 13% cheaper than baseline.
Cost per million tokens for models optimized for complex reasoning, chain-of-thought, and judgment tasks. Includes models like o1, o3-mini, DeepSeek-R1.
Cost per million tokens for the most capable general-purpose models. Includes Claude Sonnet, GPT-4o, Gemini Pro.
Cost per million tokens for high-throughput, cost-optimized models. Includes GPT-4o-mini, Gemini Flash, Claude Haiku.
Cost per million tokens for models optimized for large context windows (100K+ tokens). Used for document processing, RAG, and analysis of large codebases.
How much you pay per unit of model quality. Lower QAP means better value - more capability for your dollar.
Whether a model is over or underpriced relative to its capabilities compared to market average. Values above 1.2 suggest the model is underpriced (buy signal). Below 0.8 suggests overpriced.
Quality rating from the Chatbot Arena, where humans compare model outputs head-to-head. Higher ELO means the model wins more comparisons. Based on millions of human votes.
How fast a model is gaining or losing market share week-over-week. Positive MSV means growing adoption. Derived from OpenRouter volume data.
The price difference between frontier and budget tiers. Measures how much extra you pay for top-tier capability vs commodity.
The price difference between reasoning and frontier tiers. Measures the premium users pay for extended thinking and complex reasoning.
Cost index for a typical startup workload: heavy on coding assistance, some RAG/context, and light routing/classification.
Cost index for autonomous agent workloads: dominated by reasoning/thinking, with tool use and structured output.
Cost index for high-volume processing: bulk extraction, classification, and summarization at scale.
Weekly token velocity across tiers. Measures whether overall compute demand is accelerating or decelerating.
Tracks whether users are trading up to premium tiers or down to budget. "Premiumizing" means money moving up-market. "Commoditizing" means the opposite.