List your company

Theory of Mind reasoning on MMToM-QA

98.5Overall AccuracyUserHarness
31.4248.83566.2583.665Mar 25, 2026Apr 4, 2026Apr 14, 2026Apr 25, 2026May 5, 2026May 15, 2026May 26, 2026

Evaluation results

MethodLinks
UserHarnessBackbone=Claude-Sonnet-4.62026.0598.5——
UserHarnessBackbone=Claude-Opus-4.72026.0598.33——
UserHarnessBackbone=Qwen3-14B2026.0598.17——
UserHarnessBackbone=GPT-5.4-mini2026.0598.17——
UserHarnessBackbone=Qwen3-32B2026.0598——
UserHarnessBackbone=GPT-5.42026.0597.83——
UserHarnessBackbone=Gemini-3.1-Pro2026.0597.83——
UserHarnessBackbone=Qwen3-8B2026.0597.67——
UserHarnessBackbone=Gemini-3-Flash2026.0597.5——
UserHarnessBackbone=GPT-OSS-120B2026.0597.33——
UserHarnessBackbone=GPT-OSS-20B2026.0597——
UserHarnessBackbone=Llama-3.1-8B2026.0596.67——
AutoToMBackbone=GPT-4o2026.0583——
Ledger promptingBackbone=GPT-OSS-120B2026.0569.83——
Paradox promptingBackbone=GPT-OSS-120B2026.0567.5——
VisionToM w/ Qwen2.5-VL-7B-InstructSetting=Video Only, Nframe=82026.0366.370.762
Direct promptingBackbone=GPT-OSS-120B2026.0564.17——
Paradox promptingBackbone=GPT-5.42026.0563.33——
Ledger promptingBackbone=GPT-5.42026.0562.83——
BIP-ALM-LLaMA2Setting=Video Only, Nframe=-2026.0361.26458.3
Qwen2.5-VL-7B-Instruct TransferSetting=Video Only, Nframe=8, Transfer=true2026.0360.564.356.7
BIP-ALMSetting=reported setting2026.0556.17——
LIMPSetting=reported setting2026.0555.33——
SimToMSetting=reported setting2026.0551——
Ledger promptingBackbone=GPT-OSS-20B2026.0548.83——
Direct promptingBackbone=GPT-5.42026.0548.67——
Paradox promptingBackbone=Qwen3-32B2026.0548——
Direct promptingBackbone=GPT-OSS-20B2026.0547.33——
Ledger promptingBackbone=Llama-3.1-8B2026.0547.33——
Paradox promptingBackbone=Llama-3.1-8B2026.0547——
Ledger promptingBackbone=Qwen3-32B2026.0547——
Paradox promptingBackbone=GPT-OSS-20B2026.0546.67——
Ledger promptingBackbone=Qwen3-8B2026.0545.67——
Paradox promptingBackbone=Qwen3-14B2026.0545.17——
Ledger promptingBackbone=Qwen3-14B2026.0544.33——
Paradox promptingBackbone=GPT-5.4-mini2026.0543.5——
Direct promptingBackbone=Llama-3.1-8B2026.0543.33——
Paradox promptingBackbone=Qwen3-8B2026.0539.67——
Qwen2.5-VL-7B-InstructSetting=Video Only, Nframe=82026.0338.24729.3
Direct promptingBackbone=Qwen3-32B2026.0537.5——
Direct promptingBackbone=Qwen3-8B2026.0537.17——
Direct promptingBackbone=Qwen3-14B2026.0537.17——
Direct promptingBackbone=GPT-5.4-mini2026.0535——
Ledger promptingBackbone=GPT-5.4-mini2026.0534——