Intelligence Index vs. cost per task

Reasoning configurations listed by AA, from low to max where specified. Top left is better: smarter and cheaper. Some models are hidden by default; turn them on in the legend or use Show all.

Use the Anthropic/OpenAI/Google buttons to hide or show a whole group, or click an individual model. Click two or three points in a chart to compare them.

Intelligence Index vs. cost per task

Weighted average cost (USD) per Intelligence Index task, at API prices

Drag across the chart to zoom in; double-click empty space, Esc or Reset zoom to zoom out.

Source: Artificial Analysis, all measurements checked 17 November 2026. All 53 configurations use Intelligence Index v5.0.1; scores and task costs are not directly comparable with earlier index versions. Cost is AA’s weighted average API cost per index task, including token and cache pricing. Time is estimated weighted decode time, excluding time to first token and overhead; AA uses first-party API speed or the median across providers when unavailable. AA lists Gemini 4 Carbon at low, medium and high only. GPT-6.2 Astra and Sol are shown by default with filled markers; their predecessors use hollow markers and dashed lines. Pareto covers visible models only.

Table with all values