List your company
ResearchFormerly WizWand

Mathematical Reasoning on AIME 2025 (average@8 score)

35Average@8 ScoreDeepTool
-0.7768.51217.827.088May 28, 2026
Updated 5mo ago

Evaluation results

MethodLinks
DeepToolBackbone=Qwen3-4B, Training Stage=SFT2026.0535
ReToolBackbone=Qwen2.5-7B, Training Stage=RL-enhanced tool reasoning2026.0530
ARPO-7BBackbone=Qwen2.5-7B, Training Stage=RL-enhanced tool reasoning2026.0530
DeepToolBackbone=Qwen3-4B, Training Stage=+RL2026.0530
DeepToolBackbone=Qwen2.5-7B, Training Stage=+RL2026.0528.6
ToRLBackbone=Qwen2.5-7B, Training Stage=RL-enhanced tool reasoning2026.0526.7
DeepToolBackbone=Qwen2.5-7B, Training Stage=SFT2026.0525.8
ZeroTIR-7BBackbone=Qwen2.5-7B, Training Stage=Zero-RL2026.0525
DeepToolBackbone=Qwen2.5-7B, Training Stage=w/o thinking2026.0514.2
DeepToolBackbone=Qwen3-4B, Training Stage=w/o thinking2026.0513.3
Search-o1Backbone=Qwen2.5-7B, Training Stage=Search-based2026.0510
Search-R1Backbone=Qwen2.5-7B, Training Stage=Search-based2026.056.7
Qwen3-4B-BaseBackbone=Qwen3-4B, Training Stage=Base2026.056.3
Qwen2.5-7B-BaseBackbone=Qwen2.5-7B, Training Stage=Base2026.051.1
Qwen2.5-7B-TIRBackbone=Qwen2.5-7B, Training Stage=TIR2026.050.6