AGORA - Education: leaderboard

Metric: Accuracy (%) on the 49 education, science and academia questions (152 documents), archive-grounded agentic document reasoning with verifiable numeric answers over the domain collection of Markdown workplace documents, agent in the mini-swe-agent harness (bash tool only), temperature 1.0 with reasoning effort at the maximum each model supports; higher is better. Source: arxiv.org. Saturation forecast: Around December 2026. 8 models tracked.

Top models

#ModelScore
1Gemini 3.1 Pro (Preview) (High)59.18
2GPT-5.5 (xHigh)48.98
3DeepSeek V4 Flash (Max)46.94
4DeepSeek V4 Pro (Max)38.78
5GLM-5.134.69
6Qwen 3.5 35B A3B12.24
7Gemini 3.1 Flash Lite (High)10.2
8Qwen 3.5 9B2.04

Interactive version: theaggregate.ai/benchmark?slug=agora-education · How It Works · Data refreshed daily, snapshot 2026-09-29.