Goedel Architect
A customized system in the field of automated theorem proving, appearing as a comparison baseline on the Putnam Bench benchmark.
Timeline (last 90 days)
On Putnam Bench, when using the same model as Goedel Architect, H1 is cheaper but performs better.
The paper emphasizes that all costs should be converted to API-equivalent costs based on actual Token usage for fair comparison, because the computation per attempt differs enormously across systems (for example, one attempt of Goedel Architect allows up to 16 proof plan revisions, and one attempt of XProver Base allows 50 agent iterations).
All Facts (2)
On Putnam Bench, when using the same model as Goedel Architect, H1 is cheaper but performs better.
50%UnverifiedThe paper emphasizes that all costs should be converted to API-equivalent costs based on actual Token usage for fair comparison, because the computation per attempt differs enormously across systems (for example, one attempt of Goedel Architect allows up to 16 proof plan revisions, and one attempt of XProver Base allows 50 agent iterations).
50%