Benchmark profile

GDPval-AA

An agentic real-world work-task evaluation reported as an Elo score in DeepSeek-V4 thinking-mode evaluations.

Data verified July 29, 2026

Benchmark score on GDPval-AA — July 29, 2026

BenchLM mirrors the published score view for GDPval-AA. Claude Opus 5 leads the public snapshot at 1862 , followed by Claude Fable 5 (1747) and GPT-5.6 Sol (1735). BenchLM does not use these results to rank models overall.

1Closed

Claude Opus 5

Anthropic

claude-opus-5

1862

Overall 82.81

2Closed

Claude Fable 5

Anthropic

claude-fable-5

1747

Overall 82.75Context 1M+

3Closed

GPT-5.6 Sol

OpenAI

gpt-5-6-sol

1735

Overall 81.39Context 1M

87 modelsAgenticCurrentDisplay onlyUpdated July 29, 2026

Benchmark score table (87 models)

Score

Claude Opus 5Anthropic · Closed

1862

Claude Fable 5Anthropic · Closed

1747

GPT-5.6 SolOpenAI · Closed

1735

Kimi K3Moonshot AI · Closed

1687

Claude Sonnet 5Anthropic · Closed

1603

Claude Opus 4.8Anthropic · Closed

1593

GPT-5.6 TerraOpenAI · Closed

1583

GPT-5.6 LunaOpenAI · Closed

1582

Grok 4.5xAI · Closed

1528

GLM-5.2Z.AI · Open weight

1510

Claude Opus 4.7 (Adaptive)Anthropic · Closed

1493

GPT-5.5OpenAI · Closed

1491

Gemini 3.6 FlashGoogle · Closed

1423

GPT-5.4OpenAI · Closed

1392

MiniMax M3MiniMax · Open weight

1391

Muse Spark 1.1Meta · Closed

1375

Gemini 3.5 FlashGoogle · Closed

1345

DeepSeek V4 Pro (Max)DeepSeek · Open weight

1306

DeepSeek V4 Pro (High)DeepSeek · Open weight

1299

Qwen3.7 MaxAlibaba · Closed

1271

MiMo-V2.5-ProXiaomi · Closed

1265

GLM-5.1Z.AI · Open weight

1256

InklingThinking Machines Lab · Open weight

1237

Hy3 PreviewTencent · Open weight

1215

Hy3Tencent · Open weight

1215

DeepSeek V4 Flash (Max)DeepSeek · Open weight

1189

Kimi K2.6Moonshot AI · Open weight

1188

Kimi K2.7 CodeMoonshot AI · Open weight

1186

GPT-5.4 miniOpenAI · Closed

1169

GLM-4.7Z.AI · Open weight

1166

Nemotron 3 UltraNVIDIA · Open weight

1162

MiniMax M2.7MiniMax · Open weight

1159

DeepSeek V4 Flash (High)DeepSeek · Open weight

1147

Muse SparkMeta · Closed

1143

Qwen3.6 PlusAlibaba · Closed

1139

Gemini 3.5 Flash-LiteGoogle · Closed

1139

Qwen3.6-27BAlibaba · Open weight

1138

GPT-5.4 nanoOpenAI · Closed

1101

Grok 4.3xAI · Closed

1084

GPT-5 (high)OpenAI · Closed

1079

Qwen3.6-35B-A3BAlibaba · Open weight

1052

Step 3.7 FlashStepFun · Open weight

1017

Kimi K2.5Moonshot AI · Open weight

1003

Kimi K2.5 (Reasoning)Moonshot AI · Closed

1003

GPT-5.1OpenAI · Closed

988

Qwen3.5-122B-A10BAlibaba · Open weight

982

Gemini 3.1 ProGoogle · Closed

965

Qwen3.5 397BAlibaba · Open weight

962

Qwen3.5 397B (Reasoning)Alibaba · Open weight

962

Qwen3.7 PlusAlibaba · Closed

943

Mistral Medium 3.5 128BMistral · Open weight

933

GPT-5 miniOpenAI · Closed

932

MiMo-V2-FlashXiaomi · Open weight

838

Gemma 4 31BGoogle · Open weight

811

GPT-OSS 120BOpenAI · Open weight

802

Gemma 4 26B A4BGoogle · Open weight

770

Command A+Cohere · Open weight

718

Nemotron 3 Super 120B A12BNVIDIA · Open weight

699

Mercury 2Inception · Closed

698

Gemini 2.5 ProGoogle · Closed

672

Gemma 4 12BGoogle · Open weight

651

Gemini 3.1 Flash-LiteGoogle · Closed

647

Mistral Large 3Mistral · Closed

640

K-ExaoneLG AI Research · Closed

598

Mistral Small 4Mistral · Open weight

592

Mistral Small 4 (Reasoning)Mistral · Open weight

592

GPT-OSS 20BOpenAI · Open weight

567

Trinity-Large-PreviewArcee AI · Open weight

564

Trinity-Large-ThinkingArcee AI · Open weight

564

Ling 2.6 FlashInclusionAI · Open weight

550

GPT-4.1 miniOpenAI · Closed

508

Nemotron 3 Nano 30BNVIDIA · Open weight

492

Ministral 3 14B (Reasoning)Mistral · Open weight

487

Ministral 3 14BMistral · Open weight

487

Nemotron 3 Nano Omni 30B A3BNVIDIA · Open weight

465

Ministral 3 8B (Reasoning)Mistral · Open weight

456

Ministral 3 8BMistral · Open weight

456

Ministral 3 3B (Reasoning)Mistral · Open weight

286

Ministral 3 3BMistral · Open weight

286

GPT-4o miniOpenAI · Closed

241

DeepSeek V3DeepSeek · Open weight

231

Gemma 4 E4BGoogle · Open weight

231

Llama 4 ScoutMeta · Open weight

111

Gemma 4 E2BGoogle · Open weight

GPT-4.1 nanoOpenAI · Closed

Llama 4 MaverickMeta · Open weight

Gemma 3 27BGoogle · Open weight

-119

The published GDPval-AA snapshot places Claude Opus 5 first at 1862. The third row is 127 score units behind. The broader top-10 range is 352 score units, so the table still separates the published systems.

87 models have been evaluated on GDPval-AA. The benchmark falls in the Agentic category. This category carries a 22% weight in BenchLM.ai's overall scoring system. GDPval-AA is currently displayed for reference but excluded from the scoring formula, so it does not directly affect overall rankings.

About GDPval-AA

Year

2026

Tasks

Agentic real-world work tasks

Format

Elo

Difficulty

Professional agentic workflows

BenchLM stores GDPval-AA as a display-only provider-table row for DeepSeek-V4 because the source reports an Elo score rather than a 0-100 percentage.

DeepSeek-V4 Technical Report

BenchLM freshness & provenance

Version

GDPval-AA 2026

Refresh cadence

Quarterly

Staleness state

Current

Question availability

Public benchmark set

CurrentDisplay only

BenchLM uses freshness metadata to decide whether a benchmark should still be treated as a strong differentiator, a benchmark to watch, or a display-only reference. For the full scoring policy, see the BenchLM methodology page.

FAQ

What does GDPval-AA measure?

An agentic real-world work-task evaluation reported as an Elo score in DeepSeek-V4 thinking-mode evaluations.

Which model scores highest on GDPval-AA?

Claude Opus 5 by Anthropic currently leads with a score of 1862 on GDPval-AA.

How many models are evaluated on GDPval-AA?

87 AI models have been evaluated on GDPval-AA on BenchLM.

Compare Top Models on GDPval-AA

Claude Opus 5 vs Claude Fable 5 Claude Fable 5 vs GPT-5.6 Sol GPT-5.6 Sol vs Kimi K3 Kimi K3 vs Claude Sonnet 5

Last updated: July 29, 2026 · BenchLM version GDPval-AA 2026

Know when it’s worth switching models

The model to choose, the cheaper alternative, and the release we would wait on.

Read a sample issue

Join 2,000+ readers.

One email each week. Unsubscribe anytime.