ERNIE 5.5 Flash Base at a glance
Baidu published the weights of ERNIE 5.5 Flash Base, a 280B-parameter Mixture-of-Experts model (14B active) trained on 33.5T tokens. It is a raw pre-trained base, not a chat model. Baidu's own tests put it ahead of three similar open base models, and its licence excludes the EU, UK and South Korea.
What changed with ERNIE 5.5 Flash Base
On November 5, 2026, Baidu's ERNIE team released ERNIE 5.5 Flash Base on Hugging Face and ModelScope, in FP8 and BF16 versions. It is a text-only Mixture-of-Experts model with 280B total parameters and 14B active per token, pre-trained from scratch on 33.5 trillion tokens. Training used contexts up to 128K tokens, and Baidu says fine-tuning and inference support up to 1,048,576 tokens. It can be served with SGLang or vLLM.
This is the checkpoint at the end of pre-training, before any instruction tuning, so it completes text rather than following instructions or chatting. Baidu presents it as a starting point for companies and researchers who want to train their own version for a domain, for instruction following or for agents. A post-trained ERNIE 5.5 Flash has not been released.
Baidu compared it with three open base models of similar size, DeepSeek-V4-Flash-Base, Hy3-preview-Base and MiMo-V2.5-Base, using the same evaluation pipeline for all four. In Baidu's table, ERNIE 5.5 Flash Base scores 93.3 on MMLU (DeepSeek-V4-Flash-Base: 89.4), 78.4 on MMLU-Pro (69.2), 60.2 on GPQA (55.9), 82.3 on HumanEval+ (69.5) and 40.1 on LiveCodeBench (31.9). It scored lower than at least one of the others on ChineseSimpleQA, INCLUDE-lite, FullStackBench, HELMET-Rerank and InfiniteBench. These are Baidu's own results; as of November 7, 2026, Artificial Analysis had not published a page for the model.
The weights are free to download, but the licence is Baidu's own ERNIE 5.5 Flash Base Model License Agreement v1.0, not Apache or MIT. It applies everywhere except the European Union, the United Kingdom and South Korea. Research, education and personal use are free. Commercial use is free for products whose related annual revenue is US$50 million or less, if you send Baidu a one-time notice and show "Powered by ERNIE"; above that you need Baidu's written authorization first.
ERNIE 5.5 Flash Base in numbers
Values are copied from the linked source on the date shown. They are not IA Compa measurements.
- Context window (fine-tuning and inference)as of · Source for Context window (fine-tuning and inference): huggingface.co (opens in a new tab)
- 1M tokens
- MMLU, 5-shot (Baidu's evaluation)as of · Source for MMLU, 5-shot (Baidu's evaluation): huggingface.co (opens in a new tab)
- 93.3
- HumanEval+, pass@1 (Baidu's evaluation)as of · Source for HumanEval+, pass@1 (Baidu's evaluation): huggingface.co (opens in a new tab)
- 82.3
How it compares
Current models we have covered: released in the last six months, newest version of each line only. Each value is the model's latest recorded value. (Models)
Where to compare ERNIE 5.5 Flash Base
Independent leaderboards and official resources for comparing this model yourself. Rankings change often; each site shows its own current data.
General leaderboards
Why ERNIE 5.5 Flash Base matters for businesses
Why it matters: ERNIE 5.5 Flash Base is a large open base model you can host and fine-tune yourself, with a long context window. For agencies, it suits teams that train their own models rather than call a ready-made chatbot. Check the licence terms for your clients' locations and revenue before you build on it, and test it on your own tasks, since the benchmarks are Baidu's.
Sources for this story
- ERNIE 5.5 Flash Base (technical blog) (opens in a new tab)Official
Baidu · ernie.baidu.com
- ERNIE-5.5-Flash-Base model card and evaluation table (opens in a new tab)Official
Baidu (Hugging Face) · huggingface.co
- ERNIE 5.5 Flash Base Model License Agreement v1.0 (opens in a new tab)Official
Baidu (Hugging Face) · huggingface.co
Last verified: . IA Compa summarizes public announcements for business readers. We are not affiliated with the companies mentioned. Always confirm details, pricing, and availability with the official source.
- #llm
- #baidu
- #ernie
- #open-weights
- #mixture-of-experts
- #base-model
- #fine-tuning
