ERNIE 4.5 21B A3B Thinking
This 21B-total, 3B-active language checkpoint extends the reasoning training of ERNIE 4.5. It targets complex problems and supports function calls.
Model design and access
Baidu explicitly warns that this release uses longer thinking sequences. Its card identifies the 128K context, 28 layers and separate reasoning and tool-call parser requirements in serving examples.ERNIE-4.5-21B-A3B-Thinking official model card
Practical use
Practical guidance. Measure completion accuracy together with generated tokens. A longer reasoning trace can improve difficult tasks but also delays simple responses. Include interrupted generations in your review so that an unfinished trace is not presented as a final answer.ERNIE-4.5-21B-A3B-Thinking official model card
Specifications & access
| Detail | Value and conditions |
|---|---|
| Access | Downloadable |
| Model identifiers | ERNIE-4.5-21B-A3B-Thinking |
| Context window | 131,072 tokens Official checkpoint context window. Input and generation share this limit. ERNIE-4.5-21B-A3B-Thinking official model card |
| License | Apache 2.0ERNIE-4.5-21B-A3B-Thinking official model card |
| Documented deployment example | One 80GB GPU in the documented FastDeploy setupERNIE-4.5-21B-A3B-Thinking official model card |
Related models
- ERNIE 4.5 21B A3B
Baidu ยท ERNIE 4.5
Post-trained sparse language model with 21 billion total parameters and about three billion active per token.
Launch date not confirmedDownloadable