Atlas

Models

← All models

ERNIE 4.5 VL 424B A47B

Baidu · 2025-06-30 · 423.5B parameters

ERNIE 4.5 VL 424B A47B is Baidu's post-trained multimodal Mixture-of-Experts chat model, based on ERNIE-4.5-VL-424B-A47B-Base and released here as PaddlePaddle weights. The checkpoint contains 423,526,285,184 parameters; Baidu rounds this to 424B total and reports 47B activated per token. It accepts text, images, and video, combines a 54-layer vision-language backbone with a 32-layer vision encoder, supports switchable thinking and non-thinking modes over a 131,072-token context, and was post-trained in three supervised fine-tuning stages followed by reasoning reinforcement learning.

Benchmark scores

BenchmarkScore
BullshitBench v2 Leaderboard11.0
BuseyBench SVG1.6
SnakeBench Average Apples5.2
SnakeBench Best Apples16.0
SnakeBench Rating25.2
SnakeBench Total Apples136
SnakeBench Win Rate47.8
Loading Atlas data…