Archiboost outscored Structured AI and Nomic Agent across all three AEC-Bench scopes
On AEC-Bench, Archiboost scored 94.6% on multi-sheet drawing review, 87.5% on project-level review, and 82.5% on single-sheet review, ahead of the published Structured AI and Nomic Agent results in every scope.

We ran Archiboost on AEC-Bench on June 26, 2026. We scored 94.6% on multi-sheet drawing review, 87.5% on project-level review, and 82.5% on single-sheet review. All three scores exceeded the corresponding results published by Structured AI and Nomic.
The clearest difference showed up in project-level review, where the answer can depend on reading a drawing, a specification, and a submittal together. Archiboost finished 6.2 percentage points ahead of Structured AI and 25.5 points ahead of Nomic Agent on that work. Structured AI's results · Nomic's results
From a single sheet to a full project
Consider a door detail. On one sheet, a reviewer might check whether the title matches the assembly shown. Across a drawing set, the question changes: which plans call out that detail, and do the references lead to the right place? At the project level, the reviewer may also need to determine whether the hardware a supplier submitted meets the specified requirements.
AEC-Bench tests those different kinds of review through 196 task instances in nine categories. The drawing sets, tasks, and evaluation code are public, so teams can examine what went into each score. AEC-Bench dataset and evaluation code
| Review scope | Archiboost | Structured AI | Nomic Agent |
|---|---|---|---|
| Single sheet | 82.5% | 76.7% | 70.6% |
| Multi-sheet drawing set | 94.6% | 88.8% | 88.3% |
| Full project | 87.5% | 81.3% | 62.0% |
Archiboost scores are from our run on June 26, 2026. Competitor scores are from each company's published AEC-Bench evaluation. Structured AI's scope breakdown · Nomic's scope breakdown
The biggest task-level gain was in submittal review
A product data sheet can list several options. The reviewer has to identify which one the supplier is proposing, find the requirement that applies, and decide whether the two agree. Across the benchmark's 36 submittal-review cases, Archiboost scored 83.9%, compared with 75.0% for Structured AI. That 8.9-point margin was our largest lead over Structured AI on an individual task category. Structured AI's submittal-review results
Across all nine task categories, Archiboost scored higher than Structured AI in eight and tied on drawing navigation. The breakdown also shows where our system has the most room to improve.
| Review task | Archiboost | Structured AI |
|---|---|---|
| Find the requested drawing | 100.0% | 100.0% |
| Check the sheet index against the drawing set | 100.0% | 92.9% |
| Trace references back to a detail | 96.8% | 95.8% |
| Check whether a detail title matches the drawing | 93.3% | 86.7% |
| Verify that cross-references resolve | 91.9% | 88.2% |
| Answer technical questions about a detail | 89.7% | 85.7% |
| Identify conflicts between specifications and drawings | 86.2% | 81.3% |
| Review submittals against project requirements | 83.9% | 75.0% |
| Check whether a note points to the intended element | 63.6% | 57.1% |
The last row is worth a closer look. Archiboost earned full marks for finding the requested drawing, but scored 63.6% when it had to determine whether a note referred to the correct element. For a reviewer deciding where to spend a second look, a leader line landing between two adjacent elements deserves more scrutiny than a sheet lookup.
AEC-Bench is one of several benchmarks we're running. Stay tuned for many more results in the coming weeks.
Bring a coordination question from your own project to an Archiboost demo. Choose a submittal your team has already reviewed or a detail that appears on several sheets. We'll work through it with you so you can compare the finding with your team's decision.

