List your company

Multi-step reasoning on TriviaQA

80.09Task PerformanceLAR
54.339661.024867.7174.3952May 18, 2026

Evaluation results

MethodLinks
LARBackbone=Qwen3-8B2026.0580.09-27.1
ReActBackbone=Qwen3-8B2026.0577.84—
COTBackbone=Llama-3.1 8B-Instruct2026.0575.5—
VanillaBackbone=Llama-3.1 8B-Instruct2026.0573.63—
LARBackbone=Llama-3.1 8B-Instruct2026.0572.46-23.3
COTBackbone=Qwen3-8B2026.0569.43—
TokenSkipBackbone=Llama-3.1 8B-Instruct2026.0568.87-12
ConciseHintBackbone=Qwen3-8B2026.0568.69-12.7
VanillaBackbone=Qwen3-8B2026.0567.4—
ConciseHintBackbone=Llama-3.1 8B-Instruct2026.0567.32-13.5
ReActBackbone=Llama-3.1 8B-Instruct2026.0559.88—
ACONBackbone=Llama-3.1 8B-Instruct2026.0557.14-25.4
TokenSkipBackbone=Qwen3-8B2026.0557.02-28.7
ACONBackbone=Qwen3-8B2026.0555.33-27.9