Compact vision-language MoE model from Baidu — 28B total, 3B active.
The GA release of DeepSeek's open 1.6T-parameter MoE flagship for advanced reasoning and long-horizon agents.
No shared demos for these models yet.
No captured outputs for these models yet.
This model hasn't been benchmarked yet.
Both ERNIE 4.5 VL 28B and DeepSeek V4 Pro (0813) are capable AI models. The best choice depends on your specific use case: consider pricing, context window, speed, and which capabilities (vision, reasoning, audio) you need.
DeepSeek V4 Pro (0813) has a larger context window: ERNIE 4.5 VL 28B supports 30K tokens vs DeepSeek V4 Pro (0813) at 1M tokens.
ERNIE 4.5 VL 28B supports vision/image input, but DeepSeek V4 Pro (0813) does not.
Key differences: only ERNIE 4.5 VL 28B supports vision input; only DeepSeek V4 Pro (0813) offers extended reasoning mode. Compare full specs on this page.
If cost is your priority, choose the cheaper option. If you need the highest intelligence for complex tasks, pick the higher-scoring model. For long documents or codebases, choose the larger context window. You can try both ERNIE 4.5 VL 28B and DeepSeek V4 Pro (0813) for free on idapt.app to see which performs better for your specific needs.