AI Model Benchmarks and Rankings

Independent measurements are kept separate from Frontrumors editorial assessments. Missing results are shown as unavailable, never estimated.

Artificial Analysis snapshot checked 2026-11-01, Intelligence Index v5.0. Older Arena and Terminal-Bench measurements were not refreshed in that snapshot; their source notes remain below.

Measurement provenance and limitations

Benchmark sources and methodology

• Artificial Analysis Intelligence Index v5 (checked 1 Nov 2026): One v5 snapshot, best listed effort variant per model; effort and fallback settings are labeled. v5 launched 27 Oct and is not directly comparable with the older v4.3 scale. Older Arena and Terminal-Bench metrics were not refreshed; price/spec dates remain source-specific.
• Artificial Analysis v5 methodology announcement: Versioned methodology: nine evaluations, with a revised mix and more private test data. Do not infer model regressions from a change of index version.
• LMArena text (via AI Blogly leaderboard, 26 Oct 2026 refresh): LMArena text Elo, top ten only; models below the top ten keep their last recorded value.
• Terminal-Bench 4.0 (via Techietory snapshot): Terminal-Bench figures not on the AA leaderboard; not refreshed in this update.
• LMArena Agent arena (via BenchLeader): LMArena Agent task scores (GPT-6 Astra 14, GPT-6.1 Sol 13.6); not refreshed in this update.

Index versions and reasoning settings are part of each result. Do not compare different AA index versions as though they used the same scale. These measurements do not imply that a model leads every task.