GDPval-AA

We show this table for reference; we do not rank on it.

Data verified September 27, 202634 confirmed releases in the last 30 daysFollow model changes

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

Benchmark score on GDPval-AA — September 27, 2026

We compile the GDPval-AA rows from provider self-reports and secondary reports. Claude Opus 5 leads the table at 1862, followed by Claude Opus 5.5 (1846) and GLM-5.3-Flash (1773). We do not use these results to rank models overall.

1Proprietary

Claude Opus 5

Anthropic

1862

Overall 79.81

2Proprietary

Claude Opus 5.5

Anthropic

1846

Overall 86.98Context 1M

3Open weights

GLM-5.3-Flash

Z.AI

1773

Overall 58.01Context 1M

115 modelsAgenticCurrentDisplay onlyUpdated September 27, 2026

Benchmark score table (115 models)

Score

Claude Opus 5Anthropic · Closed

1862

Claude Opus 5.5Anthropic · Closed

1846

GLM-5.3-FlashZ.AI · Open weight

1773

GLM-5.3Z.AI · Open weight

1769

Muse Spark 1.3Meta · Closed

1754

Claude Fable 5Anthropic · Closed

1747

Claude Fable 5.1Anthropic · Closed

1735

GPT-5.6 SolOpenAI · Closed

1735

Grok 4.7xAI · Closed

1695

Hy4 previewTencent · Open weight

1678

MiMo-V2.6-ProXiaomi · Open weight

1673

Qwen3.8-Flash-NextAlibaba · Open weight

1648

Muse Spark 1.2Meta · Closed

1631

Qwen3.8 Max PreviewAlibaba · Closed

1630

Grok 4.6xAI · Closed

1605

Claude Sonnet 5Anthropic · Closed

1603

DeepSeek V4.1 FlashDeepSeek · Open weight

1600

Claude Opus 4.8Anthropic · Closed

1593

GPT-5.6 TerraOpenAI · Closed

1583

Atria Dawn PreviewShanghai Artificial Intelligence Laboratory · Open weight

1583

GPT-5.6 LunaOpenAI · Closed

1582

Step 5 PreviewStepFun · Closed

1566

Gemini 3.8 FlashGoogle · Closed

1545

GPT-6 AstraOpenAI · Closed

1542

Gemini 3.7 FlashGoogle · Closed

1525

Kimi K3Moonshot AI · Closed

1524

Qwen3.8-27BAlibaba · Open weight

1463

Grok 4.5xAI · Closed

1430

Gemini 3.6 FlashGoogle · Closed

1423

GLM-5.2Z.AI · Open weight

1418

GPT-5.5OpenAI · Closed

1396

Claude Opus 4.7 (Adaptive)Anthropic · Closed

1396

Muse Spark 1.1Meta · Closed

1375

Gemini 3.5 FlashGoogle · Closed

1345

GPT-5.4OpenAI · Closed

1307

DeepSeek V4 Pro 0813DeepSeek · Open weight

1306

MiniMax M3MiniMax · Open weight

1304

DeepSeek V4 Pro (High)DeepSeek · Open weight

1294

Apodex 1.1Apodex · Closed

1273

Apodex 1.1 MiniApodex · Open weight

1273

MiMo-V2.5-ProXiaomi · Closed

1265

Quasar 438BMultiverse Computing · Closed

1239

Inkling-SmallThinking Machines Lab · Open weight

1191

Qwen3.7 MaxAlibaba · Closed

1190

DeepSeek V4 Flash 0731DeepSeek · Open weight

1189

GLM-5.1Z.AI · Open weight

1181

InklingThinking Machines Lab · Open weight

1165

Gemini 3.5 Flash-LiteGoogle · Closed

1139

Hy3Tencent · Open weight

1136

Hy3 PreviewTencent · Open weight

1136

Kimi K2.6Moonshot AI · Open weight

1115

Kimi K2.7 CodeMoonshot AI · Open weight

1114

Ling 3.0 FlashInclusionAI · Open weight

1107

GLM-4.7Z.AI · Open weight

1096

GPT-5.4 miniOpenAI · Closed

1095

Nemotron 3 UltraNVIDIA · Open weight

1091

MiniMax M2.7MiniMax · Open weight

1087

Muse SparkMeta · Closed

1076

Qwen3.6-27BAlibaba · Open weight

1069

Qwen3.6 PlusAlibaba · Closed

1066

Ling 3.0 Flash FP8InclusionAI · Open weight

1036

GPT-5.4 nanoOpenAI · Closed

1035

Grok 4.3xAI · Closed

1018

GPT-5 (high)OpenAI · Closed

1015

Qwen3.6-35B-A3BAlibaba · Open weight

992

Step 3.7 FlashStepFun · Open weight

954

Kimi K2.5Moonshot AI · Open weight

936

Kimi K2.5 (Reasoning)Moonshot AI · Closed

936

GPT-5.1OpenAI · Closed

930

Qwen3.5-122B-A10BAlibaba · Open weight

925

Gemini 3.1 ProGoogle · Closed

904

Muse Glimmer 30BMeta · Open weight

893

Qwen3.7 PlusAlibaba · Closed

886

GPT-5 miniOpenAI · Closed

880

Mistral Medium 3.5 128BMistral · Open weight

875

Nemotron 3.5 Lightning 30B A3B NVFP4NVIDIA · Open weight

865

MiMo-V2-FlashXiaomi · Open weight

788

Gemma 4 31BGoogle · Open weight

755

GPT-OSS 120BOpenAI · Open weight

745

Gemma 4 26B A4BGoogle · Open weight

713

Command A+Cohere · Open weight

658

Granite 4.2 8BIBM · Open weight

646

Mercury 2Inception · Closed

646

Nemotron 3 Super 100BNVIDIA · Open weight

644

Nemotron 3 Super 120B A12BNVIDIA · Open weight

644

Gemini 2.5 ProGoogle · Closed

616

Gemma 4 12BGoogle · Open weight

591

Mistral Large 3Mistral · Closed

588

Ling 2.6 FlashInclusionAI · Open weight

550

K-ExaoneLG AI Research · Closed

540

Mistral Small 4Mistral · Open weight

537

Mistral Small 4 (Reasoning)Mistral · Open weight

537

GPT-OSS 20BOpenAI · Open weight

514

Trinity-Large-ThinkingArcee AI · Open weight

512

Trinity-Large-PreviewArcee AI · Open weight

512

Celeris-1Celeris · Closed

483

GPT-4.1 miniOpenAI · Closed

453

Nemotron 3 Nano 30BNVIDIA · Open weight

438

Ministral 3 14B (Reasoning)Mistral · Open weight

432

100

Ministral 3 14BMistral · Open weight

432

101

Nemotron 3 Nano Omni 30B A3BNVIDIA · Open weight

416

102

Ministral 3 8B (Reasoning)Mistral · Open weight

405

103

Ministral 3 8BMistral · Open weight

405

104

Ministral 3 3B (Reasoning)Mistral · Open weight

234

105

Ministral 3 3BMistral · Open weight

234

106

LFM2.5-2.6BLiquidAI · Open weight

202

107

GPT-4o miniOpenAI · Closed

188

108

DeepSeek V3DeepSeek · Open weight

183

109

Gemma 4 E4BGoogle · Open weight

177

110

Llama 4 ScoutMeta · Open weight

111

Ultravox v0.6 Llama 3.3 70BFixie AI · Open weight

112

Gemma 4 E2BGoogle · Open weight

113

GPT-4.1 nanoOpenAI · Closed

114

Llama 4 MaverickMeta · Open weight

-45

115

Gemma 3 27BGoogle · Open weight

-171

Among the reported GDPval-AA rows, Claude Opus 5 is first at 1862. The third row is 89 score units behind. The broader top-10 range is 184 score units, so the table still separates the published systems.

115 models have been evaluated on GDPval-AA. The benchmark falls in the Agentic category. 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

Freshness and 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.

Questions

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?

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

Compare top models on GDPval-AA

Claude Opus 5 vs Claude Opus 5.5 Claude Opus 5.5 vs GLM-5.3-Flash GLM-5.3-Flash vs GLM-5.3 GLM-5.3 vs Muse Spark 1.3

Last updated: September 27, 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.