SpacebarBench scores agents building a budget project on a whiteboard

What the research found

SpacebarBench is an October 2026 benchmark of six model configurations on fourteen consecutive requests building one customer-support vendor budget project, scored by a 100-point audit of delivered artifacts rather than a statistical model ranking.

Claims and how well the evidence supports them

1. SpacebarBench uses fourteen consecutive requests on one evolving project 97%
2. Six model configurations were evaluated 97%
3. Scores come from a 100-point audit of delivered artifacts 98%
4. Results are not a statistical model ranking 98%

Evidence

• spacebar.ai https://spacebar.ai/benchmark

Actionability rated low.