ToolUse.1 - Spurious Call Rate: leaderboard
ToolUse.1 - Spurious Call Rate. ToolUse.1 version 1.0.0 uses corpus SHA-256 17a6ef84c22648d83add6f56a03b24f896420837f4e69039c22d61cf8ad2c3e1: 114 English cases, comprising 78 tool-call and 36 no-call cases. Each model receives JSON, plain-text, and XML protocols, for 342 decisions. Runs use greedy decoding, at most 256 new tokens, no few-shot examples, thinking disabled, and batch size eight. Grading checks tool names and arguments with documented aliases, plus abstention on negative or clarification cases. Nova 1.2 is a distinct LoRA derivative of Nova 1.1. The oracle marked self_check=true validates the grader and is not a model result. Reported metric: Rate (0–1).
Metric: Rate (0–1). Source: huggingface.co. Saturation forecast: Around January 2027. 3 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3.5 0.8B (Non-reasoning) | 0.06 | #1428 (Qwen 3.5 0.8B) |
No result here: #5 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=tooluse-1-spurious-call-rate · How It Works · Data refreshed daily, snapshot 2026-10-25.