OPENLAWN FIELD NOTES · SEPTEMBER 29, 2026

GPT-6.1 Sol is the new king in this lawn-mapping head-to-head.

We gave Sol and Astra the same residential aerial, parcel context, mapping tools, medium reasoning setting and animated drawing. Sol finished faster, at a much lower reported model cost, while drawing a visibly similar lawn outline.

One propertySame imageMedium reasoning3 calls each

Actual OpenLawn head-to-head screen with aerial and lawn outlines: Astra on the left, GPT-6.1 Sol on the right — Actual saved head-to-head screen. Astra is left; Sol is right. Aerial imagery © Mapbox. The green areas are estimated lawn, and dashed lines show the parcel context.

10.1 secondsfaster visible completion41.4 s Sol vs. 51.5 s Astra · 19.6% less time

75.8%lower reported model cost$0.0285 Sol vs. $0.1178 Astra · 4.1× cost difference

54 sq ftapart in area estimate6,780 sq ft Sol vs. 6,726 sq ft Astra · 0.8% difference

01 / THE RESULT

Same property. Two live maps.

Both models completed the measurement with three model calls. Sol started drawing its first polygon at 21.6 seconds, 2.5 seconds before Astra. It reached visible completion 10.1 seconds earlier. Its reported model cost was $0.0285, compared with Astra’s $0.1178.

Measure	GPT-6 Astra	GPT-6.1 Sol
Estimated lawn	6,726 sq ft	6,780 sq ft
Visible completion	51.5 s	41.4 s
First polygon drawn	24.1 s	21.6 s
Server processing	50.8 s	40.5 s
Model calls	3	3
Reported model cost	$0.1178	$0.0285

The total square-foot estimates differed by 54 sq ft, about 0.8% of Astra’s estimate. The two maps look close overall; inspect the outlines in the image above to see where the models differ around the house and side lawn.

02 / WHAT WE CAN CONCLUDE

A strong challenger, with one test behind it.

Sol won this run on speed and reported model cost. Similar square-foot totals do not prove that either outline is accurate: a missed lawn patch and an included non-lawn patch can cancel in the total. This run has no independently surveyed or user-corrected reference outline, so we cannot assign either model an accuracy score from it.

Both requests ran concurrently; provider load, caching and network conditions can affect timing. The reported costs reflect model usage in this run, not imagery, parcel data, storage or all of OpenLawn’s operating costs. We would repeat this comparison across different properties and score the outlines against reviewed references before changing the production default.

03 / TEST NOTES

What was held constant.

This was OpenLawn Color Playground’s saved head-to-head run from September 29, 2026 at 8:21 p.m. Eastern, on one Michigan residential property. Both sides used Mapbox satellite imagery framed to the same property, whole-yard scope, the same mapping workflow and animated drawing. Each requested medium reasoning. The model IDs were openai/gpt-6-astra and openai/gpt-6.1-sol.

The image is the unaltered screenshot of the completed comparison. The private experiment archive retains the source aerial, results, event replay, timing data and recording. We omit the street address here.

Measure the lawn. Review the map.

After this case study, GPT-6.1 Sol became the default in Agent View. Users can remeasure a completed Sol result with Astra for another credit and compare both saved maps. Every aerial measurement needs a human check before it drives a quote.

Try OpenLawn free ↗Read our broader model benchmark ↗