{"@context":"https://neupai.io/schema/v0.2","@type":"StructuredNewsArticle","identity":{"article_id":"tech42_20261029_kakao-kanana-embedding-model-release","canonical_url":"https://www.tech42.co.kr/%ec%b9%b4%ec%b9%b4%ec%98%a4-ai-%ec%97%90%ec%9d%b4%ec%a0%84%ed%8a%b8%ec%9a%a9-%eb%a7%a5%eb%9d%bd-%ea%b2%80%ec%83%89-%eb%aa%a8%eb%8d%b8-%ea%b3%b5%ea%b0%9c3b%c2%b71-3b-2/?utm_source=rss&utm_medium=rss&utm_campaign=%25ec%25b9%25b4%25ec%25b9%25b4%25ec%2598%25a4-ai-%25ec%2597%2590%25ec%259d%25b4%25ec%25a0%2584%25ed%258a%25b8%25ec%259a%25a9-%25eb%25a7%25a5%25eb%259d%25bd-%25ea%25b2%2580%25ec%2583%2589-%25eb%25aa%25a8%25eb%258d%25b8-%25ea%25b3%25b5%25ea%25b0%259c3b%25c2%25b71-3b-2","ai_url":null,"publisher":{"name":"테크42","domain":"www.tech42.co.kr","type":"online"},"author":"김한수 기자","published_at":"2026-10-29T00:39:54.000Z","updated_at":null,"language":"en","article_type":"straight_news","originality":"self_produced"},"content":{"headline":"Kakao Unveils 'Contextual Search' Models for AI Agents… Two Versions, 3B and 1.3B","summary":"Kakao announced the release of two embedding models for AI agents, 'Kanana-2-embedding,' stating that it applied MRL and knowledge distillation techniques to improve search performance and operational efficiency. The company demonstrated superior Korean-language performance compared to comparable global models and applied an open license allowing commercial use.","topics":["AI","embedding models","Kakao","open source"],"geography":["KR"],"entities":[{"name":"Kakao","canonical_id":"corp:kr:kakao","type":"company","role_in_article":"primary_subject","metadata":{"ticker":"372180.KS","parent":null}},{"name":"Kanana-2-3b-embedding","canonical_id":"product:kr:kanana-2-3b-embedding","type":"product","role_in_article":"mentioned","metadata":{"ticker":null,"parent":"corp:kr:kakao"}},{"name":"Kanana-2-1.3b-embedding","canonical_id":"product:kr:kanana-2-1-3b-embedding","type":"product","role_in_article":"mentioned","metadata":{"ticker":null,"parent":"corp:kr:kakao"}},{"name":"Noh Byung-seok","canonical_id":"person:kr:nobong-seok","type":"person","role_in_article":"quoted","metadata":{"ticker":null,"parent":"corp:kr:kakao"}},{"name":"Hugging Face","canonical_id":"corp:us:huggingface","type":"company","role_in_article":"mentioned","metadata":{"ticker":null,"parent":null}}],"claims":[{"id":"c1","statement":"On the 29th, Kakao uploaded the 3-billion-parameter 'Kanana-2-3b-embedding' and the 1.3-billion-parameter 'Kanana-2-1.3b-embedding' to Hugging Face","as_of":"2026-10","as_of_explicit":true,"as_of_raw":"the 29th","source_type":"company_disclosure","comparison":null,"type":"fact","figures":{"value":2,"unit":"count","approximate":false,"converted":null},"expiry_hint":null,"insight":null},{"id":"c2","statement":"The Kanana-2-3b-embedding model has 3 billion parameters","as_of":"2026-10","as_of_explicit":true,"as_of_raw":"the 29th","source_type":"company_disclosure","comparison":null,"type":"fact","figures":{"value":3000000000,"unit":"parameters","approximate":false,"converted":null},"expiry_hint":null,"insight":null},{"id":"c3","statement":"The Kanana-2-1.3b-embedding model has 1.3 billion parameters","as_of":"2026-10","as_of_explicit":true,"as_of_raw":"the 29th","source_type":"company_disclosure","comparison":null,"type":"fact","figures":{"value":1300000000,"unit":"parameters","approximate":false,"converted":null},"expiry_hint":null,"insight":null},{"id":"c4","statement":"Kakao previously unveiled four small language models in July","as_of":"2026-07","as_of_explicit":true,"as_of_raw":"last July","source_type":"company_disclosure","comparison":null,"type":"fact","figures":{"value":4,"unit":"count","approximate":false,"converted":null},"expiry_hint":null,"insight":null},{"id":"c5","statement":"In an evaluation comparing them against llama-nemotron-embed-1b-v2 and Qwen3-Embedding-4B, the released models achieved the highest scores on Korean-language items","as_of":"2026-10","as_of_explicit":false,"as_of_raw":"October 2026","source_type":"company_disclosure","comparison":"competitor_models","type":"fact","figures":null,"expiry_hint":null,"insight":null},{"id":"c6","statement":"Compared to the previous Kanana embedding series, search performance improved, and memory usage and inference speed showed efficiency gains of more than tens of times","as_of":"2026-10","as_of_explicit":false,"as_of_raw":"October 2026","source_type":"company_disclosure","comparison":"previous_version","type":"fact","figures":{"value":null,"unit":"multiple","approximate":true,"converted":null},"expiry_hint":null,"insight":null},{"id":"c7","statement":"Kakao plans to apply larger language models to embedding training going forward and continue research on knowledge distillation","as_of":"2026-10","as_of_explicit":false,"as_of_raw":"October 2026","source_type":"company_plan","comparison":null,"type":"future_plan","figures":null,"expiry_hint":null,"insight":null}],"ai_emotional_context":{"valence":0.4,"arousal":0.3,"primary_emotions":[{"emotion":"optimistic","intensity":0.5}],"secondary_emotions":[{"emotion":"calm","intensity":0.3}],"emotional_triggers":[{"claim_id":"c5","emotion":"optimistic","reason":"significant_positive_metric"}]},"image":{"url":"https://onjblseywainslkhvkav.supabase.co/storage/v1/object/public/article-images/35305fbd-6559-4ba5-b14e-816fa7b7f2b4.jpg","alt":"카카오 Kanana-2-embedding 모델이 한국어·영어 성능 평가에서 경쟁 모델들보다 높은 점수를 기록한 막대그래프.","caption":null,"source":"first_img","alt_status":"auto"}},"provenance":{"source_chain":["primary_reporting"],"original_source_url":null,"related_articles":[]},"temporal":{"freshness":"breaking","next_update_expected":null},"access":{"license":"neupai_standard","attribution_required":true,"structured_data":"free","full_text_available":false,"full_text_access":null}}