PaddlePaddle's PP-OCRv6 model family for Hugging Face covers 50 languages with parameter counts from 1.5 million to 34.5 million.
PaddlePaddle has published PP-OCRv6 on Hugging Face, an optical character recognition model family that supports 50 languages. According to the announcement, the models range from 1.5 million to 34.5 million parameters.
The announcement centers on the model family's language coverage and parameter range. With only one source, there are no differing accounts to compare; the details above come directly from the Hugging Face blog post.
Meta's release of Muse Spark Open narrows the gap between American and Chinese open-weight models, but its restrictive license may complicate adoption.
The new checkpoint is now live across DeepSeek's platform and API, with the company claiming faster, shorter outputs and a new search API.
StepFun's open-weight Step 5 Preview brings a sparse 600B-parameter MoE model with 27B active parameters and 1M-token context for agentic workloads.
A credit manager is using Codex to build tools and answer questions about loan valuations and credit agreements faster.