BrickAGI benchmark

Models

One card per (model, protocol) baseline. Ranked rows are current, complete, and task-set-pinned; diagnostic rows remain visible for auditability.

gpt-5.5

openai
scaffold-assembly-v1
CaSS 35.0%
coverage 55% cost $5.80

gpt-5.5

openai
scaffold-v1
CaSS 45.0%
coverage 0% cost $4.54 partial breaker DNR ×3

gemini-3-pro-preview

google
raw-v1
CaSS 39.6%
coverage 0% cost $4.20 partial breaker DNR ×5

gpt-5.5

openai
raw-v1
CaSS 35.0%
coverage 5% cost $4.90 legacy breaker