AEC-Bench - Note Callout Accuracy: leaderboard
Metric: Mean reward (0-100) from the task verifier on the 14 note-callout-accuracy instances (verifying whether callout text correctly describes the referenced element) built from real public-sector construction documents, one trial per instance, where the agent runs in its coding-agent harness (Codex for GPT, Claude Code for Claude) with Bash and command-line PDF tools; higher is better. Source: arxiv.org. Saturation forecast: Around 2033. 4 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Sonnet 4.6 | 42.9 | #85 |
| 2 | GPT-5.4 | 28.6 | #76 |
| 3 | GPT-5.2 | 28.6 | #105 |
| 4 | Claude Opus 4.6 | 0 | #60 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=aec-bench-note-callout-accuracy · How It Works · Data refreshed daily, snapshot 2026-10-11.