GTA-Workflow: leaderboard

Root success on GTA-Workflow, the open-ended tier of GTA-2 (2026): 132 long-horizon productivity workflows from real user queries with deployed tools and multimodal inputs, run end to end in a ReAct agent framework and judged by GPT-5.2 against recursive checkpoint rubrics: the share of tasks whose final deliverable scores above 7 of 10 at the root checkpoint. Percent; higher is better.

Metric: Root success rate (%): share of tasks whose root checkpoint score exceeds 7 of 10 on GTA-Workflow, the open-ended long-horizon workflow tier of GTA-2 (132 real-user tasks with deployed tools and multimodal inputs), run end to end in the default Lagent ReAct framework; deliverables judged by GPT-5.2 against recursive checkpoint rubrics; higher is better. Source: arxiv.org. Saturation forecast: Around November 2027. 13 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Gemini 2.5 Pro14.39#276
2GPT-511.36#178
3Qwen 3 235B A22B10.61#628
4Llama 4 Scout10.61#1818
5Grok 49.85#319
6Claude Sonnet 4.59.09#261
7DeepSeek V3.29.09#419
8Kimi K28.33#443
9Llama 3.1 70B Instruct0.76#1571
10Llama 3.2 3B Instruct0.76#7241
11Qwen 3 30B A3B0.76#1156
12Llama 3.1 8B Instruct0#4600
13Qwen 3 8B0#1873

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=gta-workflow · How It Works · Data refreshed daily, snapshot 2026-11-21.