Mystery 'Carbon 4' model tops Artificial Analysis knowledge benchmark at 54, above Opus 5.5
YashasGunderia · x · 2026-10-31
A circulating Artificial Analysis AA-Omniscience v1.1 chart — which rewards correct answers, penalizes hallucinations, and doesn't punish refusals — shows an unreleased model labeled "Carbon 4" (no publicly available API) topping the board at 54, ahead of Claude Opus 5.5 (48), Opus 5 (46), GPT-5.1 (42), and Grok 4.1 (42). The poster calls it an interesting jump for Google and can't wait to try it, though the model's maker is not officially confirmed.