Vendor
Baidu has released 5 models: 4 language models and 1 specialist model.
Baidu is a Chinese internet company headquartered in Beijing, founded by Robin Li and Eric Xu on January 18, 2000. Robin Li is CEO. The company runs the search engine of the same name and develops the ERNIE family of AI models.
Baidu unveiled the Ernie Bot chatbot in 2023 and released ERNIE 4.0 in October 2023. On March 16, 2025 it followed with the foundation model ERNIE 4.5 and the reasoning model ERNIE X1. The ERNIE 4.5 family was open-sourced on June 30, 2025. On Hugging Face, Baidu also provides models such as ERNIE-Image, Unlimited-OCR and Qianfan-VL.
ERNIE 5.5 Flash Base, with open weights, appeared on November 5, 2026. The Mixture-of-Experts model has 280 billion parameters, 14 billion of them active per token. It was pre-trained on 33.5 trillion tokens and is available as a base checkpoint without post-training under the ERNIE 5.5 Flash Base Model License Agreement v1.0.
Every model there ever was. Models with a profile lead to every benchmark, price and source.
| ERNIE 5.5 Flash Base | Nov 5, 2026 | Active | n/a | n/a | n/a | 1.05M | n/a |
| ERNIE 4.5 21B A3B | Jun 30, 2025 | Active | n/a | n/a | n/a | 131K | 45 |
| ERNIE 4.5 300B A47B | Jun 30, 2025 | Active | n/a | n/a | n/a | 131K | 39 |
| ERNIE 4.5 VL 424B A47B | Jun 30, 2025 | Active | n/a | $0.42 / $1.25 | n/a | 131K | 25 |
| Unlimited OCR 2 | Oct 27, 2026 | Open weights | n/a | 58 |
Language model · Nov 5, 2026
ERNIE 5.5 Flash Base is a base model Baidu released on November 5, 2026. It is the checkpoint from the end of pretraining, before any mid- or post-training. Baidu positions it as a starting point for domain adaptation, instruction tuning, and reinforcement learning research. The model handles text only.
Specialist model · Oct 27, 2026
Unlimited OCR 2 parses multi-page documents with an approach that avoids keeping the full visual history in GPU KV cache. Baidu publishes separate detailed results for MPDocBench, MarathonDocBench, and OmniDocBench v1.6. The dataset card lists about 18,000 pages across 100 MarathonDocBench documents of 100 to 300 pages. The technical report lists 103 to 300 pages for its 100 documents, with a median of 178 pages. The benchmarks cover different tasks and report separate metrics.
Language model · Jun 30, 2025
ERNIE 4.5 21B A3B is the smallest text MoE of Baidu's ERNIE 4.5 series, with 21 billion total and 3 billion active parameters per token. Six of 64 text experts are activated per token.
Language model · Jun 30, 2025
ERNIE 4.5 300B A47B is the large text MoE of the ERNIE 4.5 series, with 300 billion total and 47 billion active parameters per token. Eight of 64 text experts are activated per token.
Language model · Jun 30, 2025
ERNIE 4.5 VL 424B A47B combines image understanding with a heterogeneous mixture-of-experts architecture. The open model accepts text and images, activates 47 of 424 billion parameters, and uses 54 layers with a 131,072-token context.
By release date. Grey: retired models.
2 releases since January 2026, by release date.
Prices per 1M tokens for the current language models, according to Baidu or the provider linked in each model profile.
| ERNIE 4.5 VL 424B A47B | $0.42 | $1.25 | 131K |
The directory lists 5 Baidu models: 4 language models and 1 specialist model. It counts every model Baidu ever released, including retired ones. 5 are currently available, the newest being ERNIE 5.5 Flash Base, Unlimited OCR 2 and ERNIE 4.5 21B A3B.
ERNIE 5.5 Flash Base, released on Nov 5, 2026. The timeline below shows every release by date.
Through the API per 1M tokens (input / output): ERNIE 4.5 VL 424B A47B $0.42 / $1.25. The prices of all 1 current language models are in the price table.
Yes, 5 of the 5 models have open weights. Each model profile lists how much memory it needs.
The largest is ERNIE 5.5 Flash Base with 1.05 M tokens (≈ 1,570 A4 pages). Every model's value is in the model list.
Data as of Nov 6, 2026