AI Intelligence per Dollar — Frontier Model Benchmark Efficiency (2020–2026)
Tracks how much AI capability a dollar buys across six years of frontier model releases. Combines MMLU, HumanEval, MATH, BIG-Bench Hard, and GPQA Diamond scores into a normalised Composite Capability Index, then divides by published API inference cost to produce an Intelligence-per-Dollar ratio. Covers GPT-3 (2020) through GPT-5, Claude Sonnet 4.6, and Gemini 2.5 Pro (2026).
