Capability rankings · 2026-10-02

Rankings

Five capability rankings use one independent Artificial Analysis evaluation snapshot. Each card shows the original benchmark scores.

Browse models

Same evaluator and benchmark version. Tied scores share a rank; missing scores receive no rank. Overall ranking →

19 evaluated models · %

Coding

64%

Terminal-Bench 4.0 in the Artificial Analysis evaluation harness.

Claude Sonnet 5.5 64% 2

Claude Opus 5.5 60% 3

GPT-6 Astra 59%

Artificial Analysis · Checked 2026-10-02

19 evaluated models · %

Reasoning

61%

Humanity's Last Exam as evaluated by Artificial Analysis; the source configuration is preserved.

Claude Opus 5.5 61% 2

Claude Fable 5.1 59% 3

Gemini 4 Argon 57%

Artificial Analysis · Checked 2026-10-02

19 evaluated models · Elo

Research

1846

GDPval-AA v2.1 professional knowledge-work Elo. This is a task lens, not an assessment of every research skill.

Claude Opus 5.5 1846 2

Claude Sonnet 5.5 1844 3

Claude Fable 5.1 1735

Artificial Analysis · Checked 2026-10-02

19 evaluated models · %

Multimodal

31%

GDP.pdf professional document understanding, including visual tables, charts and diagrams; not an audio/video score.

GPT-6 Astra 31% 1

GPT-6.1 Sol 31% 3

Claude Fable 5.1 26%

Artificial Analysis · Checked 2026-10-02

19 evaluated models · %

Agentic

78%

AutomationBench-AA multi-step business workflows in the Artificial Analysis harness.

Gemini 4 Argon 78% 2

Claude Sonnet 5.5 71% 3

Claude Opus 5.5 70%

Artificial Analysis · Checked 2026-10-02

Ranking basis

Scores come from the same independent evaluation suite, with one published configuration per model. Overall averages the five normalized capability indices on a 0–100 scale.

Raw evidence remains separate

Model profiles preserve original units, reasoning settings, source links and historical results. Read the methodology →