Hi Team
I’m writing on behalf of Humanfia team to submit our result for inclusion on the PutnamBench leaderboard.
Our system successfully solved 669 out of 672 problems in PutnamBench using an agentic proof-generation loop built with PolyArch/humanize. The average cost per problem is about $44.6 using GPT-5.5
All components of our implementation will be open source, including the agent framework, generated Lean proofs, verification results, and associated code. The complete result can therefore be independently inspected and reproduced.
I have sent the email once before. Could you take some time to check the proof? We would like to hear your feedback and update our results on the leaderboard.
Bests,
Ligeng Zhu
Hi Team
I’m writing on behalf of Humanfia team to submit our result for inclusion on the PutnamBench leaderboard.
Our system successfully solved 669 out of 672 problems in PutnamBench using an agentic proof-generation loop built with PolyArch/humanize. The average cost per problem is about $44.6 using GPT-5.5
All components of our implementation will be open source, including the agent framework, generated Lean proofs, verification results, and associated code. The complete result can therefore be independently inspected and reproduced.
I have sent the email once before. Could you take some time to check the proof? We would like to hear your feedback and update our results on the leaderboard.
Bests,
Ligeng Zhu