KAIST releases KORMo-Web-11B, an open Korean-English web agent
Original title: KORMo-VL/KORMo-Web-11B — new model trending #29 on Hugging Face
The 11B Apache-2.0 model keeps language and vision skills and reports 30.1% Korean Online-Mind2Web success.
KAIST MLP Lab released KORMo-Web-11B, built from its from-scratch KORMo-10B language model and extended through KORMo-VL. The lab says the single model handles text, images, and browser GUI tasks in Korean and English after training on more than 10 million web samples and more than 100,000 interaction trajectories. Weights are under Apache-2.0. On 612 live-website tasks, the model card reports 29.0% English and 30.1% Korean success.
Unconfirmed: Success rates are the lab’s own evaluation; independent verification is not stated.
- The evaluation used 300 English and 312 Korean Online-Mind2Web tasks.
- The card says the model’s thinking traces are produced only in Korean.
- KORMo-Web-4B and 9B are described as stronger on web-only tasks.
- The vision stage uses a SigLIP encoder with the KORMo language backbone.
- Organizations
- KAIST MLP Lab
- Countries
- South Korea
- AI models
- KORMo-Web-11BKORMo-10BKORMo-VL
From the original article
9 min read at huggingface.coimage-text-to-text · 277 downloads · 98 likes · license apache-2. 0
By KORMo-VL · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.