Liveta/agent-migration-golden-suite
Viewer • Updated • 40 • 57
Note 40-case golden suite: does the candidate model still call the right tools? Ready to run with EvalShift.
Note Measures "negative flips" when a model update gets previously-correct cases wrong
Note Regression testing between a current and candidate LLM; GPT-4o showed regressions vs GPT-3.5
Note "Model drifting": prompts tuned for one model degrade on another
Note Standard function-calling eval set
View the Berkeley Function‑Calling Leaderboard
Note Live tool-calling scores by model
Note Agent trajectories from 17 models on the same tasks, useful to see how models differ
Note MCP tool-call eval: selection, params, values