This edition captures the Artificial Analysis (AA) Intelligence Index v4.3.2 table on October 18, 2026. It uses that index alone for ordering and keeps the highest-scoring tested configuration per model version. Effort and fallback settings remain in each name. Separate model versions stay separate. This is a dated capture, not a final result for the whole month.

Reading the results

Claude Opus 5.5 (max with fallback) leads this selection at a displayed 58. Claude Sonnet 5.5 (max with fallback) and Muse Wonder Full (max) show 56. Claude Fable 5.1 (max with fallback), GPT-6 Astra (max) and Gemini 4 Argon (high) show 53. Equal displayed integers do not establish equality at full precision or statistical significance. Source order breaks display ties without asserting superiority.

What does a score above 1600 mean?

Check the complete metric name. AA Intelligence Index is a composite index. AA’s GDPval-AA and AA-Briefcase can instead use Elo scales; GDPval-AA v2.1 currently anchors DeepSeek V4.1 Flash (max) at 1600. Arena text preference ratings use a separate pairwise-comparison scale and baseline. These ratings cannot be compared with 58 index points, a historical 59.6 LLM Stats score, or a 93% benchmark result. No simple division converts one into another.

Why use AA for the main table?

Our editorial recommendation is to fix one source and metric for a monthly capability series, then consult task evaluations, human preference and cost separately. The current AA categories are Agents 30%, Coding 20%, Scientific Reasoning 20% and General 30%. Its scope is primarily English text. Chinese writing, vision, speech and actual business performance need separate evaluation.

Configuration and cost

Effort settings and fallback behavior are part of the tested configuration. Cost per Task is AA’s reported USD per benchmark task, not an API rate per million tokens or a subscription price. Missing values remain unreported, not zero. Deduplication prevents repeated effort variants from dominating the list; row positions are therefore not the raw ranks on AA’s all-configuration table. These are not necessarily vendors’ default settings. Preview models and models without displayed pricing may be included; inclusion does not establish universal availability.