List your company
ResearchFormerly WizWand

Accuracy on RACE Reading Comprehension

93.3Accuracy (RACE)Mistral-Large-3 675B-Base-2512
22.26840.70959.1577.591Apr 17, 2025Jun 27, 2025Sep 6, 2025Nov 16, 2025Jan 26, 2026Apr 7, 2026Jun 17, 2026
Updated 4mo ago

Evaluation results

MethodLinks
Mistral-Large-3 675B-Base-2512shots=0-shot2026.0693.3
DeepSeek-V3.2 Exp-Baseshots=0-shot2026.0693.21
Nemotron-3-Ultra 550B-A55B-Baseshots=0-shot2026.0692.15
GLM-4.5 Baseshots=0-shot2026.0692.15
Kimi-K2 Baseshots=0-shot2026.0691.96
Llama-3.2Size=1.2B2025.0437.8
CLIMBSize=950M2025.0436.65
TinyLlamaSize=364.62025.0436.46
Qwen2.5Size=490M2025.0434.93
SmolLMSize=360M2025.0434.74
AMD-OLMoSize=1.2B2025.0434.64
MoE baselineModel Family=MoE2026.0634.6
2B constant-widthModel Family=Dense2026.0634.3
MoE > <formerModel Family=MoE2026.0634.3
2B > <formerModel Family=Dense2026.0633.4
CLIMBSize=350M2025.0433.39
RegMix-DProxy models=512, Mode=Online2026.0632.06
RegMix-DProxy models=512, Mode=Offline2026.0631.83
RegMix-DProxy models=128, Mode=Offline2026.0631.21
DoReMi2026.0631.07
RegMix-DProxy models=128, Mode=Online2026.0630.91
RegMix2026.0630.75
Human2026.0629.87
MIRParameters=1.4B, Training Data=DCLM, Training Tokens=100M, Evaluation=Zero-shot2026.0627.66
Regularized BaselineParameters=1.4B, Training Data=DCLM, Training Tokens=100M, Evaluation=Zero-shot2026.0626.89
Random GuessParameters=1.4B, Training Data=DCLM, Training Tokens=100M, Evaluation=Zero-shot2026.0625