benchmarks.site

Public leaderboard · Agents, tool use & computer use

GDPval-AA

Tests AI on real office-style professional tasks across many occupations, judged head-to-head.

Leaderboard

Top 3 right now

  1. 1stClaude Opus 5.5 (Max, Default Fallback)1866
  2. 2ndKimi K3.1 Pro1861
  3. 3rdMuse Wonder Full (Max)1857

What it measures

Real work tasks from 44 occupations across the 9 largest US GDP sectors, where models produce file deliverables; Artificial Analysis runs an agentic version reported as an Elo-style rating.

How to read the numbers

Elo rating from pairwise comparisons (higher is better) — higher is better.

Results

25 rows · higher is better (Elo)CSV
1Claude Opus 5.5 (Max, Default Fallback)1866Sep 2026
2Kimi K3.1 Pro1861Oct 2026
3Muse Wonder Full (Max)1857Oct 2026
4Claude Sonnet 5.5 (Max, Default Fallback)1838Sep 2026
5Claude Opus 5.5 (Xhigh, Default Fallback)1836Sep 2026
6Claude Fable 5.2 (Max, Default Fallback)1831Nov 2026
7Qwen4 Max1809Nov 2026
8Muse Wonder (Max)1770Oct 2026
9Claude Fable 5.1 (Max, Default Fallback)1765Sep 2026
10Claude Fable 5.2 (Xhigh, Default Fallback)1751Nov 2026
11DeepSeek V4.1 Flash 1017 (Max)1746Oct 2026
12Claude Sonnet 5.5 (Xhigh, Default Fallback)1743Sep 2026
13Claude Fable 5.1 (Xhigh, Default Fallback)1741Sep 2026
14Gemini 4 Carbon (High)1733Oct 2026
15Claude Opus 5 (Max)1729Jul 2026
16Claude Fable 5.2 (High, Default Fallback)1718Nov 2026
17Grok 4.7 (High)1718Sep 2026
18Grok 4.7 (Xhigh)1717Sep 2026
19Claude Opus 5.5 (High, Default Fallback)1709Sep 2026
20Claude Opus 5 (Xhigh)1695Jul 2026
21MiMo-V2.6-Pro1692Sep 2026
22Muse Spark 1.3 (Max)1681Sep 2026
23Qwen3.8 Max (0902)1674Sep 2026
24Grok 4.6 (Xhigh)1658Aug 2026
25Ling 3.1 Flash 10191655Oct 2026

Note: GDPval-AA v2.1, top 25 of 319 models. Site terms of use not reviewed in depth.

Source: Artificial Analysis · artificialanalysis.ai · retrieved 2026-11-17 · license: not stated (site has Terms of Use)

Markdown · More Agents, tool use & computer use