ERNIE 5.5 Flash Base
Baidu released the weights of ERNIE-5.5-Flash-Base, a sparse mixture-of-experts base model with 280B total parameters and 14B active per token.
Baidu published the model at the end of pre-training so developers can fine-tune it for their own domains, reasoning or agent uses. Of its 43 layers, 37 attend over a sequence compressed 128x and six use MLA-DSA layers, which Baidu says keeps long-context performance strong. Baidu compared it with DeepSeek V4 Flash Base, Hy3 preview Base and MiMo V2.5 Base using the same evaluation harness.
- Date
- Thursday 5 November 2026
- Lab
- Baidu
- Kind
- open-weights
- Access
- open weights
Figures
| Measure | Value | Measured by |
|---|---|---|
| Total parameters | 280B sparse MoE | company |
| Active parameters | 14B per token | company |
Baidu's release post does not state a license. The benchmark scores appear in images and tables in the post.
Sources
- ernie.baidu.com/release/ernie-5-5-flash-base/
- pandaily.com/baidu-ernie-5-5-flash-base-open-source-280b-moe-14b-active-base-model
- gigazine.net/gsc_news/en/20261106-ernie-5-5-flash-base/
This record was checked against its sources on 5 November 2026. How we check