BenchLeader
Menu
Best LLMsChoose a modelCoding plansLeaderboardsModelsBenchmarksProvidersReleases8

GDPval-AA v2.1

As of 29 Oct 2026, the top score on GDPval-AA v2.1 is 1866 by Claude Opus 5.5 (Anthropic), ahead of Muse Wonder Full at 1857, across 169 model configurations with a published result (Artificial Analysis, result dated 22 Sept 2026). GDPval asks models to produce real work products across 44 occupations. Artificial Analysis gives the model shell and web access in an agentic loop to produce them. GDPval-AA v2.1 is shown for reference only and does not count toward the BenchLeader Index.

Key facts (as of 29 Oct 2026)

  • Claude Opus 5.5 holds the top GDPval-AA v2.1 score at 1866 (Artificial Analysis, 22 Sept 2026).
  • Muse Wonder Full is second on GDPval-AA v2.1 at 1857 (Artificial Analysis, 12 Oct 2026).
  • GDPval-AA v2.1: Artificial Analysis' run of OpenAI's GDPval professional tasks with shell and web access in an agentic loop, rated as an Elo from blind pairwise comparisons of the work products (Artificial Analysis).
  • 169 model configurations have a published GDPval-AA v2.1 result on BenchLeader (BenchLeader, 29 Oct 2026).
  • GDPval-AA v2.1 does not feed the BenchLeader Index; it is shown for reference (BenchLeader methodology).

Artificial Analysis' run of OpenAI's GDPval professional tasks with shell and web access in an agentic loop, rated as an Elo from blind pairwise comparisons of the work products.

Published by
Artificial Analysis
Category
Agents & tools
Index weight
Reference only
Models
169
Data as of
29 Oct 2026
Newest result
28 Oct 2026

Source: Artificial Analysis (artificialanalysis.ai). Data taken from the public leaderboard.

Not comparable with the earlier test of the same name, whose final scores stay on GDPval-AA v2.1, normalised score.

What the test looks like

GDPval asks models to produce real work products across 44 occupations. Artificial Analysis gives the model shell and web access in an agentic loop to produce them.

How it is scored

An Elo from blind pairwise comparisons between models' work products, so a rating only means something relative to the others on the board.

What to keep in mind

Replaced the normalised percentage in late October 2026; the two are not comparable. Pairwise ratings move as models are added. The Epoch-mirrored official results are counted in the index.

Top 15
  1. 1Claude Opus 5.5 (max)1866
  2. 2Muse Wonder Full (max)1857
  3. 3Kimi K3.1 Pro1846
  4. 4Claude Sonnet 5.5 (max)1838
  5. 5Claude Opus 5.5 (xhigh)1837
  6. 6Muse Wonder (max)1770
  7. 7Claude Fable 5.1 (max)1764
  8. 8DeepSeek V4.1 Flash 1017 (max)1746
  9. 9Claude Fable 5.1 (xhigh)1740
  10. 10Claude Sonnet 5.5 (xhigh)1735
  11. 11Gemini 4 Carbon (high)1733
  12. 12Grok 4.7 (xhigh)1722
  13. 13Grok 4.7 (high)1717
  14. 14Claude Opus 5.5 (high)1713
  15. 15MiMo-V2.6-Pro1691
169 of 169
#
1Claude Opus 5.5max186669.22026-09-22
2Muse Wonder Fullmax185762.22026-10-12
3NewKimi K3.1 Pro184661.32026-10-23
4Claude Sonnet 5.5max183865.02026-09-28
5Claude Opus 5.5xhigh183766.02026-09-22
6Muse Wondermax177062.22026-10-12
7Claude Fable 5.1max176467.02026-09-01
8DeepSeek V4.1 Flash 1017max174659.32026-10-17
9Claude Fable 5.1xhigh174064.92026-09-01
10Claude Sonnet 5.5xhigh173563.22026-09-28
11NewGemini 4 Carbonhigh173365.32026-10-28
12Grok 4.7xhigh172258.12026-09-21
13Grok 4.7high171757.02026-09-21
14Claude Opus 5.5high171365.62026-09-22
15MiMo-V2.6-Pro169157.22026-09-21
16Qwen3.8 Maxmax167358.82026-09-02
17GLM 5.3max165858.92026-08-18
18Ling 3.1 Flash165350.32026-10-22
19GLM 5.3 Flash164957.22026-08-26
20Claude Fable 5.1high164563.62026-09-01
21Qwen3.8-Flash-Next163754.72026-08-26
22Gemini 4 Argonhigh163165.92026-09-30
23Claude Haiku 5.5max162856.72026-10-07
24Qwen3.8 2.4T A95B161652.82026-08-12
25MiMo-V2.6-Flash161555.02026-09-21
26GPT-6.1 Solmax160867.22026-09-29
27Step 5 Preview159454.82026-09-18
28Claude Opus 5.5medium159361.72026-09-22
29Grok 4.7low159255.42026-09-21
30GPT-6 Astramax158467.72026-09-03
31GPT-6 Astraxhigh156966.62026-09-03
32Claude Sonnet 5.5high155858.12026-09-28
33Deepseek v4 Flash Visionmax1555–2026-08-21
34Claude Fable 5.1medium155562.02026-09-01
35GPT-6.1 Solxhigh155364.32026-09-29
36GPT-6 Astrahigh153766.82026-09-03
37Claude Haiku 5.5xhigh152156.82026-10-07
38GPT-6.1 Solhigh152164.02026-09-29
39GPT-6 Astramedium149864.92026-09-03
40Claude Fable 5.1low147858.82026-09-01
41NewGPT-6.1 Terramax147664.92026-10-26
42MiniMax M3.1 Flash146652.92026-10-19
43GPT-6.1 Solmedium145963.32026-09-29
44DeepSeek V4 Promax145756.82026-08-13
45Gemini 3.8 Flashhigh144561.92026-09-02
46GPT-6 Lunamax144556.62026-09-22
47Gemini 3.8 Flashmedium143260.32026-09-02
48Qwen3.8 27Bxhigh142752.22026-08-14
49Claude Haiku 5.5high142555.52026-10-07
50NewGPT-6.1 Terraxhigh141560.32026-10-26
51GPT-6 Astralow140362.32026-09-03
52Muse Spark Openxhigh140153.22026-10-15
53Qwen3.8 27Bmedium138949.92026-08-14
54GPT-6 Lunaxhigh137351.82026-09-22
55Qwen3.8 27Blow137148.82026-08-14
56K2 Horizon 375B A23B136649.02026-09-03
57NewGPT-6.1 Terrahigh135360.02026-10-26
58GPT-6 Lunahigh135250.92026-09-22
59Agnes 2.5 Pro Beta1336–2026-08-26
60JT-4.1 Flash 236B A21Bthinking1336–2026-07-09

More agents & tools benchmarks · All LLM benchmarks

Related benchmarks

Top head-to-heads on GDPval-AA v2.1

Cite as: BenchLeader, “GDPval-AA v2.1 leaderboard”, data as of 29 Oct 2026, https://www.benchleader.com/benchmarks/aa_gdpval_elo.

GDPval-AA v2.1: questions

What does GDPval-AA v2.1 measure?
GDPval asks models to produce real work products across 44 occupations. Artificial Analysis gives the model shell and web access in an agentic loop to produce them. Scores are reported in Elo-style ratings from pairwise comparisons; higher is better.
What is the top score on GDPval-AA v2.1?
As of 29 Oct 2026, the highest score on GDPval-AA v2.1 is 1866 by Claude Opus 5.5 (result dated 22 Sept 2026), ahead of Muse Wonder Full at 1857. Scores come from Artificial Analysis and are re-read daily.
Which AI model leads GDPval-AA v2.1?
Claude Opus 5.5 leads GDPval-AA v2.1 with 1866 as of 29 Oct 2026, ahead of Muse Wonder Full at 1857 (result dated 22 Sept 2026).
How many models have GDPval-AA v2.1 results?
169 model configurations have a GDPval-AA v2.1 result on BenchLeader, all taken from Artificial Analysis.
Who runs GDPval-AA v2.1 and how often is it updated?
GDPval-AA v2.1 is published by Artificial Analysis. BenchLeader re-reads the published results once a day and records the date each result was published.
Does GDPval-AA v2.1 count toward the BenchLeader Index?
No. GDPval-AA v2.1 is shown for reference but left out of the composite index.