The Pulse
ONESTRUCTION Ranks 20 AI Models on Construction Tasks
ONESTRUCTION published a construction-focused AI leaderboard comparing 20 models across five work categories. Its October 2026 results show near-ties at the top overall, alongside substantial differences between categories.
AI.info Team · Written with AI · fact-checked ·
ONESTRUCTION’s new Construction AI Leaderboard puts a limit on the appeal of a single winner: Claude Opus 5.5 and Gemini 4 Argon finished almost level overall, while category leaders differed in three of five construction tasks. The company’s October 2026 results show that general-purpose commercial models led the aggregate ranking, but that lead did not translate into the top result across every task. ONESTRUCTION published the website on Nov. 4, with assessments of 20 commercial and publicly available models.
Five construction tasks, not one general score
The leaderboard compares models in five categories: creating BIM information requirements, understanding drawings and documents, understanding BIM data, editing BIM data, and construction knowledge such as estimating. ONESTRUCTION publishes category rankings to show where a model performs well or poorly and to help users choose a model for a particular use.
The evaluations use five named benchmarks. Ishigaki-IDS-Bench contains 166 questions on creating BIM information requirements; AEC-Bench covers 196 drawing-and-document tasks; and IFC-Bench-V2 tests BIM-data understanding with 514 questions. BIM-Edit covers 324 BIM-editing tasks, while CEQuest contains 134 questions on construction knowledge, including estimating.
ONESTRUCTION says it ran all 20 models using the same procedures on an evaluation platform it built. The company is developing Ishigaki, a construction-domain generative AI foundation model, and says its leaderboard draws on evaluation knowledge gained through that work. ONESTRUCTION was selected for the third and fourth phases of GENIAC, a domestic generative AI development project promoted by Japan’s Ministry of Economy, Trade and Industry and NEDO.
Overall leaders do not win every category
In the October 2026 evaluation, Claude Opus 5.5 scored 73.7 overall and Gemini 4 Argon scored 73.6. General-purpose large commercial models occupied the upper end of the overall ranking, but the top model varied across three of the five categories. The results therefore make the overall ranking an incomplete guide to performance on any one construction task.
Model scores clustered closely in construction knowledge: all 20 scored at least 85 points. The gaps were wider in drawing and document understanding and in BIM-data understanding. In those categories, leading scores stayed in the 60s and 70s, while some smaller publicly available models scored in the single digits.
Creating BIM information requirements proved difficult even for general-purpose models. Claude Opus 5.5 scored 63.0 in that category, the lowest leading score among the five categories. ONESTRUCTION says it used Ishigaki-IDS-Bench, a benchmark proposed in its paper accepted at the international conference CIKM 2026, for that assessment.
More models and a BIM-generation test are planned
The current rankings reflect the models and evaluation conditions in place when ONESTRUCTION published them. The company cautions that the results do not assess the technical capabilities of the organizations that develop or provide the models.
ONESTRUCTION plans to add results when new models become available, and says it will publish a sixth category on generating BIM from drawings using BIMGenBench, currently described as a provisional name. It also plans to add other models in its Ishigaki series. For now, the board covers five categories and 20 models, with no single overall leader taking the top position in every kind of work.