Tool use / function calling
StepTool
Superseded baseline#7 of 55 most-superseded
Superseded — cited as a baseline and beaten by newer methods
2 papers critique it · 1 beat it on benchmarks
What papers say
Verbatim critique sentences, each from a paper that cites StepTool as a baseline.
StepTool~yu2024steptool scores each step with GPT-4 during PPO, but the external judge can only score calls that were made, not penalize calls that should not have been, and its knowledge may diverge from the student's.
“the acquisition of process rewards heavily relies on GPT-based annotations”
Beaten on benchmarks
Head-to-head results where a newer method reports beating StepTool. Values are copied from the source paper's tables — verify against the cited paper.
CodeTool beats StepTool
69.75 vs 44.02
SoPR (Solvable Pass Rate) · [Qwen2.5-Coder-7B]
CodeTool: Enhancing Programmatic Tool Invocation of LLMs via Process Supervision
What to use instead
Recent methods in the same sub-problem, not yet superseded in the knowledge base — arXiv benchmark leaders, not vetted production recommendations.