Intelligence for your money

Methodology

Score. The vertical axis is the Artificial Analysis Intelligence Index (v4.3.2), a composite of reasoning, knowledge, maths and coding benchmarks. Scores refresh daily from their API.

Cost. The horizontal axis is the measured cost of one benchmark task in US dollars: input, cache, reasoning and answer tokens, weighted across the benchmark suite. It is a real run, not list price per token. Two models with the same token price can sit far apart here when one thinks much longer. Your own tasks and subscription billing will differ.

Log scale. Costs span three orders of magnitude, so each gridline step multiplies the price. On a linear axis everything under a dollar would collapse into one corner.

Effort.Reasoning models are tested at several effort settings. Each line joins one model's settings, cheapest to most expensive, so you can see where extra thinking stops paying for itself.

Only published pairs.A point appears once both a published score and a measured cost exist for that exact setting. Missing settings are never estimated. New models are flagged automatically and added once their cost is checked. Anthropic results include the benchmark's default fallback behaviour. Older models are kept for comparison.

Costs last checked 2026-09-30.