Opus 4.6 tops the Intelligence Index but cost $2,486 to run, more than GPT-5.2 xhigh, at unchanged pricing. The new adaptive thinking and effort controls replace thinking budgets.
Key takeaways · AI-distilled
In February 2026, Artificial Analysis found Opus 4.6 non-thinking mode scored 49% on TerminalBench versus 46% for adaptive thinking. More reasoning did not improve every benchmarkA standard public test set for comparing AI models — the shared scoreboards behind every "model X beats model Y" claim.Full definition → in the report.
Artificial Analysis estimated $2,486 to run the Intelligence Index with Opus 4.6 adaptive max in February 2026. That estimate excluded cached-input discounts, limiting its use as a prediction of a deployed application's bill.
The February 2026 Opus 4.6 report recorded gains in several reasoning evaluations alongside regressions in AA-LCR and IFBench. A higher aggregate Intelligence Index score did not imply improvement on every task.
Terms in this piece · Glossary
token — The chunk of text a model reads and writes in — roughly three-quarters of a word — and the unit AI usage is billed in.
context window — The maximum amount of text a model can consider at once — its working memory for the current conversation or task.
benchmark — A standard public test set for comparing AI models — the shared scoreboards behind every "model X beats model Y" claim.
