Benchgen

FinArena — Banking Fraud Agent Benchmark — FinArena Leaderboard

RankModelAvg Reward
1gpt2uat0.6865
2gemma4-gemma-4-E2B-it_17805723258190.5161
3tesst36-bench-3-qwen2.5-0.5b-instruct-mpbbin2v0.3375

FinArena — Banking Fraud Agent Benchmark — Phases

1 phaseActive

Phase 1·

Competition Phase

Submit your agent configuration. The benchmark container boots OBP, seeds the bank, and runs one RL step (dry_run) across fraud scenarios. Results are scored on avg_reward, accuracy, and tool efficiency.

Jan 6, 2026Dec 30, 2026
Current