ERNIE-5.5-Flash

Baidu🇨🇳 CNMajor lab· Oct 21, 2026

Textopen weights

ERNIE-5.5-Flash-Base is a sparse MoE language model with 280B total and 14B activated parameters, pre-trained on 33.5T tokens. It is the checkpoint before mid- and post-training and serves as a starting point for domain adaptation, instruction tuning and reinforcement learning research.

Summary of the model card

License: other
Modalities: text → text

2 variants

Base				HF
Base BF16				HF

Sources

HuggingFace Paper

Deutsch

ERNIE-5.5-Flash-Base ist ein MoE-Sprachmodell mit 280 Milliarden Gesamtparametern, von denen 14 Milliarden pro Token aktiv sind, vortrainiert auf 33,5 Billionen Tokens. Es ist der Checkpoint vor Mid- und Post-Training und dient als Ausgangspunkt fuer Domaenenanpassung, Instruction Tuning und Reinforcement-Learning-Forschung.

Benchmarks

Benchmark	Domain	Value	Evidence
LongBench v2vendor number	long-context	44.5 % correct	90
LiveCodeBenchvendor number	coding	31.9 % pass@1	82
SimpleQAvendor number	knowledge	29.1 % correct	81
HumanEvalvendor number	coding	69.5 % pass@1	57
RepoBenchvendor number	coding	49.1 % exact/edit	55
MGSMvendor number	multilingual	84.6 % correct	44
MMLUvendor number	knowledge	89.4 % correct	39

Vendor-reported numbers are marked and count with factor 0.7 in rankings. The evidence score says how much a benchmark still tells you today.

Provenance

Detected: Nov 6, 2026
First source: huggingface
Description: Summary of the model card
Editorially reviewed: —