September 30, 2026
Maximum performance costs more: Claude Opus 5.5, $5.98 per task
On Sep 22, Claude Opus 5.5 scored 58 points in Artificial Analysis's Intelligence Index at an average cost of $5.98 per task. max mode gave the model the index's highest score. A year earlier, GPT-5 mini in high mode scored 17 points at $0.05 per task.
Artificial Analysis
@artificialanlys
Artificial Analysis's Pareto frontier for Intelligence Index versus cost per task has shifted notably over the past 18 months. Just a year ago, the strongest-performing model was GPT-5 mini in high mode: 17 points on version 4.3 of the Intelligence Index at around $0.05 per task. Last week, Claude Opus 5.5 set a new record: 58 points at around $5.98 per task.
· 6.3K views
Over the year, the top score rose from 17 to 58 points, while the average cost per task rose from $0.05 to $5.98. These are different models and modes: GPT-5 mini was tested in high, Claude Opus 5.5 in max.
Switching to the API. Anthropic says to specify the `claude-opus-5-5` model and select an effort level. You cannot disable the legacy thinking settings: the API will return a 400 error.
When moving from Opus 5 without an effort setting, the default will change from high to medium.
