benchmarks/tasks/tau_bench/default.yaml

task: tau_bench
version: 2.0

# Official leaderboard defaults
model: gpt-4.1-mini-2025-04-14
user_llm: gpt-4.1-2025-04-14
temperature: 0.0
max_tokens: 2048
max_steps: 200
max_errors: 10
seed: 300
k: 4
task_split: test
domain: airline

# ACE settings
ace:
  epochs: 1
  max_refinement_rounds: 3
  batch_reflect: false
  trace_limit: 500