Benchmarks
SKY-001 CAD and BIM benchmarks
Skyfall AI publishes SKY-001 results for 3D CAD and, separately, for BIM construction planning. This page records those figures. It is not a test run by sky-001.org.
What the comparison is
On the product page, the headline speed and cost numbers are versus GPT-6 Astra. The IoU charts name more models. CAD and BIM are different tasks, so a CAD percentage should not be read as a BIM score. Intersection over Union (IoU) here is Skyfall’s overlap score for the generated geometry against their reference: higher is better, and a one-point gap is not the same kind of difference as a fifty-fold gap in time.
3D CAD
The page headline says 3.9 seconds and $0.005 per CAD case for SKY-001, against 199 seconds and $0.75 for GPT-6 Astra. The IoU chart on the same page, read on 21 October 2026, lists:
| Model | IoU |
|---|---|
| SKY-001 | 79.7% |
| GPT-6 Astra | 79.1% |
| Claude Opus 5.5 | 69.4% |
| Gemini 3.8 Flash | 54.7% |
| Qwen 3.8 Max | 34.4% |
| Kimi K3 | 9.5% |
| Grok 4.7 | No completed CAD cases shown |
BIM
The BIM headline says 1.22 seconds and $0.0024 per plan for SKY-001, against 144.6 seconds and $0.446 for GPT-6 Astra. The BIM IoU chart is a different list. Gemini 3.8 Flash at 42.3% is the highest non-SKY bar we could read, which matches the page’s “best LLM” BIM line. Astra is 40.6% on that chart.
| Model | IoU |
|---|---|
| SKY-001 | 43.6% |
| Gemini 3.8 Flash | 42.3% |
| GPT-6 Astra | 40.6% |
| Claude Opus 5.5 | 34.4% |
| Kimi K3 | 33.7% |
| Qwen 3.8 Max | 28.8% |
| Grok 4.7 | 28.1% |
Headline pair, CAD and BIM together
| Metric | SKY-001 | GPT-6 Astra |
|---|---|---|
| 3D CAD IoU (higher is better) | 79.7% | 79.1% |
| 3D CAD time per case | 3.9 s | 199 s |
| 3D CAD cost per case | $0.005 | $0.75 |
| BIM IoU (higher is better) | 43.6% | 40.6% |
| BIM time per plan | 1.22 s | 144.6 s |
| BIM cost per plan | $0.0024 | $0.446 |
Source: Skyfall AI, skyfall.ai/sky-001, checked 21 October 2026, alongside the 19 October 2026 introduction. Time and cost headlines compare SKY-001 with GPT-6 Astra (CAD 3.9 s vs 199 s, $0.005 vs $0.75; BIM 1.22 s vs 144.6 s, $0.0024 vs $0.446). On the same page the CAD IoU chart lists SKY-001 at 79.7% and GPT-6 Astra at 79.1%, which is also the “best LLM” CAD figure in the headline. The BIM IoU chart lists SKY-001 at 43.6%, Gemini 3.8 Flash at 42.3%, and GPT-6 Astra at 40.6%. The 42.3% headline is Gemini on that chart, not Astra. These are vendor-reported results. sky-001.org did not run the tests.
Limits
- sky-001.org did not receive the prompts, the reference solids, or the scoring code.
- Time and cost for models other than SKY-001 and GPT-6 Astra were not printed as stable headline numbers on the pass we checked, so they are not repeated here.
- Grok 4.7 is marked as having completed no CAD cases on Skyfall’s chart. That is their note, not a finding from this site.
- A vendor chart can change without notice. The last check of the product page was 21 October 2026.
- These dollar figures are benchmark costs in Skyfall’s write-up. They are not a price for using this website, and this website does not sell model calls.
Sources and update history
- 21 October 2026: IoU bars and time/cost headlines checked against the published product page and the 19 October 2026 introduction.
How those tasks are described sits on the SKY-001 CAD and BIM workflow. Whether anyone outside Skyfall can run the model is covered on the access and waitlist page.