Agentic-OPD Study - Tool Use - OlympiadBench: leaderboard
Agentic-OPD v1 appendix unmodified Qwen2.5-3B-Instruct Base. Tool Use native OlympiadBench score0–100, interactionlimit16, existingSearchR1/ReTool/AgentGymRL evaluation. No trainedGRPO/OPD row or average of unlike WebShop score/success metrics.
Metric: Score (%). Source: arxiv.org. 1 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen2.5-3B-Instruct | 17.9 |
Interactive version: theaggregate.ai/benchmark?slug=agentic-opd-study-tool-use-olympiadbench · How It Works · Data refreshed daily, snapshot 2026-11-12.