Understand this story
Kakao has open-sourced two versions of its self-developed image recognition model Kanana-2-vision-encoder: a medium version with 426 million parameters and a tiny version with 98 million parameters, both now available on Hugging Face. According to a report by The Elec on October 22, 2026, Kakao said the two models achieved the highest performance compared with Google SigLIP 2, Meta Perception Encoder, Apple AIMv2 and Alibaba Qwen3.5-ViT across 42 benchmarks in nine major categories, including general visual question answering, English-Korean OCR, document understanding and object localization; the tiny version surpassed competitors with more than 400 million parameters on some benchmarks, with particularly notable gaps in Korean recognition, document understanding and object localization. The report also said that using the self-developed Kanana-2 language model during training improved training efficiency by about 4 times, and improvements to the high-resolution image processing architecture reduced latency by more than 60%. The models are released under the Kanana Open License, permitting research, development and commercial use. There are currently no further developments.
AI-generated from public reports
Latest development · Report
Kakao Open-Sources Kanana-2-vision-encoder Vision Encoders in 426 Million and 98 Million Parameter VersionsDevelopments
1 confirmed developments
- Source:The Elec
Reports and discussion timeline
Browse reports and discussion by time. Their evidence levels remain distinct.
Oct 22
The Elec
Kakao Open-Sources Kanana-2-vision-encoder Vision Encoders in 426 Million and 98 Million Parameter Versions
Kakao has open-sourced two versions of its self-developed image recognition model Kanana-2-vision-encoder: a medium version with 426 million parameters and a tiny version with 98 million parameters, both now available on Hugging Face. Kakao says that across 42 benchmarks spanning nine major categories, including general visual question answering, English-Korean OCR, document understanding and object grounding, the two models achieved the highest performance compared with Google SigLIP 2, Meta Perception Encoder, Apple AIMv2 and Alibaba Qwen3.5-ViT. The tiny version outperforms competitors with more than 400 million parameters on some benchmarks, with particularly notable gaps in Korean recognition, document understanding and object grounding. During training, the use of the self-developed Kanana-2 language model improved training efficiency by about 4x, and architectural improvements for high-resolution image processing reduced latency by more than 60%; the models are released under the Kanana Open License, permitting research, development and commercial use.