ERNIE 4.5 21B A3B
Built by Baidu · China · research.baidu.com
The ERNIE 4.5 series of open-source models adopts a Mixture-of-Experts (MoE) architecture, representing an innovative multimodal heterogeneous model structure. It achieves cross-modal knowledge fusion through a parameter-sharing mechanism while retaining dedicated parameter spaces for individual modalities. This architecture is particularly well-suited for the continuous pre-training paradigm from large language models to multimodal models, significantly enhancing multimodal understanding capabilities while maintaining or even improving performance in text-based tasks. The models are efficiently trained, inferred, and deployed using the PaddlePaddle deep learning framework. During the pre-training of large language models, the Model FLOPs Utilization (MFU) reaches 47%. Experimental results demonstrate that this series of models achieves state-of-the-art (SOTA) performance across multiple text and multimodal benchmarks, with particularly outstanding results in instruction following, world knowledge memorizatio
How to use
Just add the suffix. Hard questions still come back from this model; only easy ones drop to something cheaper, and nothing pricier than it gets used.
model: "baidu/ernie-4.5-21b-a3b:auto"
Sets no behavior — whatever your key already stores stays in effect.
Without the suffix, the bare name always goes to this model — no routing, so nothing saved.
Context
Max output
Input $/1M
Output $/1M
Features
Providers
NovitaProviders
Price, latency, and uptime per provider serving this model. Warp tries them in order of how each host has just been behaving, moving to the next on failure. Click a row for regions and data policies.
Loading provider metrics…
Performance
Measured latency (lower is better) and throughput (higher is better) per provider.
Speed
Loading provider metrics…
Parameter support
What Warp actually does with each parameter when you call this model. The answer differs by serving provider and by API surface.