Physical AI Bench Leaderboard

Physical AI Bench (PAI-Bench) is a comprehensive benchmark suite for evaluating physical AI generation and understanding across diverse scenarios including autonomous vehicles, robotics, industrial spaces, and everyday ego-centric environments.


nvidia/Cosmos-Predict2.5-2B	82.1	86.7	77.6	94.4	68.7	86.9	89.7	84.4	91.6	91.4	93.1	99.2	51.9	69.8	21.7	97.0	96.9


Veo-3	82.1	86.7	77.6	94.4	68.7	86.9	89.7	84.4	91.6	91.4	93.1	99.2	51.9	69.8	21.7	97.0	96.9
nvidia/Cosmos-Predict2.5-2B	81.0	84.0	77.9	94.1	66.1	80.8	87.8	81.4	93.9	92.5	94.2	99.1	52.4	70.8	20.1	96.6	97.4
Wan-AI/Wan2.2-I2V-A14B	80.6	84.1	77.2	93.2	66.3	81.7	89.2	82.1	91.8	91.6	93.7	98.3	51.2	69.6	20.4	96.0	96.6
Wan-AI/Wan2.2-TI2V-5B	80.4	83.4	77.4	93.1	65.2	79.3	88.4	83.0	91.5	91.8	93.7	98.8	51.9	69.9	20.3	95.9	96.7
Wan-AI/Wan2.1-I2V-14B-720P	79.7	82.7	76.8	90.6	66.9	80.1	89.7	80.1	88.7	90.0	93.1	98.1	51.5	70.1	20.4	95.2	96.0
MAGI/MAGI-1-24B	78.5	80.5	76.5	90.6	61.8	73.5	84.1	79.8	87.7	90.0	92.4	99.0	50.2	64.2	21.4	96.8	97.9
THUDM/CogVideoX1.5-5B-I2V	78.3	80.1	76.6	89.1	59.7	73.0	84.4	79.2	91.8	91.6	93.9	98.5	50.0	66.5	21.2	95.0	96.1
THUDM/CogVideoX-5B-I2V	77.9	79.5	76.3	87.7	58.0	74.0	84.4	79.0	90.2	91.4	93.4	98.0	51.2	64.6	21.3	94.1	95.9
Lightricks/LTX-Video-13B	77.9	78.4	77.4	88.9	55.3	70.1	82.7	78.3	90.1	90.6	93.5	99.0	53.5	69.5	21.4	95.7	96.0
Tencent/HunyuanVideo-I2V	77.4	76.8	78.0	87.4	56.3	67.7	83.0	75.5	88.2	94.3	95.3	99.5	52.1	65.2	21.5	98.6	97.6
MAGI/MAGI-1-4.5B	76.9	77.4	76.3	87.5	56.3	71.6	79.8	76.0	88.9	92.1	93.3	99.0	50.4	61.8	21.6	94.5	98.1
Lightricks/LTX-Video-2B	76.9	76.6	77.1	87.3	53.6	67.1	81.5	77.1	87.6	89.2	92.7	98.7	53.2	71.3	21.1	95.0	95.9
Doubiiu/DynamiCrafter_1024	69.7	65.6	73.7	75.2	43.4	55.0	72.5	64.1	83.8	91.1	92.5	94.9	51.5	68.0	21.2	84.5	86.2