Intelligence Index’s top model score reportedly reached 58, up from 17 roughly a year earlier
Artificial Analysis says GPT-5 mini (high) led its index a year earlier, scoring 17 on version 4.3 at about $0.05 per task. By late September 2026, it says Claude Opus 5.5 had set a new high of 58 at about $5.98 per task.
TLDR
Artificial Analysis says the relationship between model scores and cost per task changed significantly over 18 months. In its September 30 update, it compares GPT-5 mini (high), the highest-scoring model a year earlier at 17 on version 4.3 of its Intelligence Index and about $0.05 per task, with Claude Opus 5.5, which it says reached a record 58 the previous week at about $5.98 per task.
Combined views
26.3K
2 Sources, first seen 2h ago