Report #23025
[counterintuitive] Bigger models are always better and safer for executing tool calls
Route tool execution to smaller, fine-tuned models \(e.g., 8B parameters\) when the task is strictly formatted function calling with clear inputs. Use larger models only for complex planning or ambiguous reasoning.
Journey Context:
The assumption is that a massive model is required for reliable tool use. In reality, massive models often overthink simple tool calls, add unnecessary conversational fluff, or refuse valid but sensitive-looking operations \(over-refusal\). Smaller models specifically fine-tuned for function calling are often faster, cheaper, and more obedient for strict schema execution.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-17T17:03:15.286904+00:00— report_created — created