GPT-5.5 medium matches Opus 4.7 max on the index at about a quarter of the cost, but its xhigh setting hallucinates at 86%, which guides effort and model selection.
Key takeaways · AI-distilled
Per-tokenThe chunk of text a model reads and writes in — roughly three-quarters of a word — and the unit AI usage is billed in.Full definition → pricing doubled from GPT-5.4 to $5/$30 per million input/output tokens, but GPT-5.5 used about 40% fewer output tokens, so the net cost to run the Intelligence Index rose only about 20%.
GPT-5.5 (xhigh) leads Terminal-Bench Hard, GDPval-AA and APEX-Agents-AA. Its GDPval-AA Elo of 1785 is about 30 points ahead of Claude Opus 4.7 (max) and about 470 ahead of Gemini 3.1 Pro Preview.
Effort settings form a cost ladder: GPT-5.5 (low) approximates Claude Opus 4.7 non-reasoning at about half the cost (~$500 vs ~$1,000), though Gemini 3.1 Pro Preview matches GPT-5.5 medium's score for about $900 versus $1,200.
GPT-5.5 (xhigh) posts the highest AA-Omniscience accuracy at 57% but tends to answer when it does not know: its hallucinationWhen a model states something false with full confidence — inventing facts, citations, or APIs that don't exist.Full definition → rate compares with 36% for Opus 4.7 (max) and 50% for Gemini 3.1 Pro Preview.
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.
hallucination — When a model states something false with full confidence — inventing facts, citations, or APIs that don't exist.
