Compare AI models and reasoning levels by their scores on the Artificial Analysis Intelligence Index and their estimated cost per benchmark task. Choose a starting setting to see what changes when you select another model or reasoning level.
Reading the chart
In Finder, choose a core model and reasoning level. Auto neighbours finds nearby trade-offs; Select neighbours lets you pin exact model and reasoning settings through Edit comparisons above the chart. Toggle reasoning pills in any model row to select multiple settings together. The selected setting sits at zero: alternatives are positioned by their change in index score horizontally and cost vertically. Click another model to make it the reference. In Select neighbours, the previous core stays in the comparison set. Select one or more lab pills to change the comparison pool; use the model selection below the chart to include or exclude model families. Nearby recommendations update for that pool.
Each arrow shows rounded percentage changes in index score (🧠) and cost ($). Hover for the model, lab, reasoning level, source values and precise changes. Solid arrows stay within a lab; dashed arrows cross labs and use the destination lab’s colour. The cost axis compresses large changes; use the printed percentages for the actual cost comparison. The label controls reduce context text when the graph is busy; selected and comparison labels stay on.
Price–score connects all measured reasoning levels for each selected model on a chart with a logarithmic cost axis. Use its model pills to select one or more curves. Reasoning compares index scores and task costs across measured effort settings. Only measured settings are shown; Sonnet and Opus use Default Fallback measurements.
What the numbers mean
The Intelligence Index combines results from several evaluations. A percentage change is a relative change in that aggregate score, not a percentage change in task accuracy. A 5% index change has no universal conversion into improvement on your own work.
Cost is the estimated US dollar cost per Intelligence Index benchmark task, including input, cache, reasoning and answer tokens. It is not subscription usage or a quote for your task. See the Artificial Analysis methodology and the chart’s model sources for details.
The chart uses an Artificial Analysis snapshot from 26 October 2026, not a live feed. It includes 15 model families and 48 settings. Two settings, Gemini 3.8 Flash Low and Muse Wonder Full Maximum, have scores but no reported cost; price comparisons exclude them.