Current ranking basis: Artificial Analysis Intelligence Index v4.3.2, checked 2026-10-02.
Source records are labeled as official, third-party, paper, leaderboard, community, or unknown.
Production claims should prefer exact model-version sources and should separate official self-reported results from third-party evaluations.
Overall restores the five-axis arithmetic average: coding, reasoning, research, multimodal and agentic have equal 20% weights. Each axis is normalized to 0–100 before averaging; raw percentages and Elo are not mixed directly. The AA Intelligence Index remains a separate benchmark result.
One published configuration is selected per exact model: normally its highest published reasoning setting, with the setting and fallbacks shown. Scores are not selected independently per test. Ties in rounded public scores share a rank; release dates do not affect scores.
Benchmark result rows identify exact model versions, source type, verification status, update date, and notes about harness or methodology differences.
Checked dates identify our source review, not the evaluator's execution date. Provider-reported results, community transcriptions and historical suites remain available but do not feed the current independent ranking.
Benchmark results are useful signals, but they do not fully represent every real-world use case. Prompting, tools, sampling, model version, and evaluation harnesses can all change results.
Maximum effort can cost much more than default effort. Small rounded differences do not establish significance. Missing evidence is neither a zero nor a low rank.
Equal-weight average of five normalized capability indices: coding, reasoning, research, multimodal and agentic.
Unit: 0–100
Terminal-Bench 4.0 in the Artificial Analysis evaluation harness.
Unit: %
Humanity's Last Exam as evaluated by Artificial Analysis; the source configuration is preserved.
Unit: %
GDPval-AA v2.1 professional knowledge-work Elo. This is a task lens, not an assessment of every research skill.
Unit: Elo
GDP.pdf professional document understanding, including visual tables, charts and diagrams; not an audio/video score.
Unit: %
AutomationBench-AA multi-step business workflows in the Artificial Analysis harness.
Unit: %
Each axis = raw benchmark score ÷ the highest score for that benchmark in this snapshot × 100, rounded to one decimal. Overall = (coding + reasoning + research + multimodal + agentic) ÷ 5, rounded to one decimal. Missing axes remain empty and produce no overall score; a measured zero is included.
Indices are relative to the selected cohort. The overall score is calculated by gpt.college and is not an independent benchmark result or the raw AA Intelligence Index. Updating the cohort can change normalized scores.
Read the evaluator's methodology →