ONESTRUCTION Releases Construction AI Leaderboard Ranking 20 Models
ONESTRUCTION released the Construction AI Leaderboard on November 4, 2026, a site that ranks generative AI models on construction work.
1 source 1h ago
Reporting from 1 source: ASCII.jp.
Why it matters
The leaderboard splits a single overall score into five task categories, and the split is the finding: cost-estimation ability is near-uniform across 20 models while drawing and BIM-data comprehension separates them, so procurement choices hinge on category rather than rank.
Key Facts
- ONESTRUCTION released the Construction AI Leaderboard on November 4, 2026, evaluating 20 commercial and open generative AI models across five construction work categories.
- Claude Opus 5.5 scored 73.7 points in the overall ranking as of October 2026, with Gemini 4 Argon at 73.6.
- The top model differed in three of the five categories on the Construction AI Leaderboard.
- The IDS category uses Ishigaki-IDS-Bench with 166 questions, drawn from an ONESTRUCTION paper accepted at CIKM 2026.
The Tottori City company evaluates 20 commercial and open models across five categories, from reading drawings and documents to editing BIM data and cost estimation. In results as of October 2026, Claude Opus 5.5 led the overall ranking at 73.7 points, with Gemini 4 Argon at 73.6. The top model differed in three of the five categories. All 20 models scored 85 points or higher on construction knowledge such as cost estimation, while drawing comprehension and BIM data understanding topped out in the 60 to 70 point range, with some small open models in single digits. A sixth category, BIM generation from drawings, is listed as coming soon. ONESTRUCTION is developing the Ishigaki construction-domain model through the third and fourth terms of GENIAC, a project run by METI and NEDO.
The IDS category uses Ishigaki-IDS-Bench, a benchmark proposed in an ONESTRUCTION paper accepted at CIKM 2026, and carries 166 questions. The other published categories are AEC-Bench for drawing and document understanding (196 tasks), IFC-Bench-V2 for BIM data understanding (514 questions), BIM-Edit for BIM data editing (324 tasks), and CEQuest for construction knowledge including cost estimation (134 questions). All models are run through the same procedure on a common platform built by ONESTRUCTION.
The company says the rankings describe models at release under its own evaluation conditions and do not rate the technical capability of the organizations that develop them. ONESTRUCTION plans to add results as new models appear, and to post evaluations of its own Ishigaki series. The same day, ONESTRUCTION opened a waiting list for Ishigaki-BIM, a beta AI agent that generates BIM models from natural-language requirements and drawings, aimed at apartment and office buildings.
- Creation of BIM information requirements (IDS): Ishigaki-IDS-Bench, 166 questions, benchmark from a paper accepted at CIKM 2026
- Understanding of drawings and documents: AEC-Bench, 196 tasks
- Understanding of BIM data: IFC-Bench-V2, 514 questions
- Editing of BIM data: BIM-Edit, 324 tasks
- Construction knowledge such as cost estimation: CEQuest, 134 questions
- Ability to generate BIM from drawings: BIMGenBench, listed as coming soon
Written with AI by Yomimono from the 1 cited source below, including Japanese-language reporting where cited. See our editorial policy.
Add Yomimono as a preferred source on Google