<?xml version="1.0" encoding="UTF-8"?>
<urlset xmlns="http://www.sitemaps.org/schemas/sitemap/0.9" xmlns:news="http://www.google.com/schemas/sitemap-news/0.9">
<url><loc>https://news.contentslab.ai/2026/10/20/sakana-qd-rec/</loc><news:news><news:publication><news:name>CONTENTSLAB NEWS</news:name><news:language>ko</news:language></news:publication><news:publication_date>2026-10-20T22:30:00+09:00</news:publication_date><news:title>'무엇을'보다 '왜'를 추천 — Sakana AI, 추천 이유를 진화시키는 QD-Rec 공개</news:title></news:news></url>
<url><loc>https://news.contentslab.ai/2026/10/20/ant-ming-3-tts/</loc><news:news><news:publication><news:name>CONTENTSLAB NEWS</news:name><news:language>ko</news:language></news:publication><news:publication_date>2026-10-20T19:50:00+09:00</news:publication_date><news:title>한국어도 되는 0.5B 오픈 음성 모델 Ming-3.0-tts — MIT 라이선스, 목소리 복제 · 감정 · 웃음 태그까지</news:title></news:news></url>
<url><loc>https://news.contentslab.ai/2026/10/20/vals-index-v22/</loc><news:news><news:publication><news:name>CONTENTSLAB NEWS</news:name><news:language>ko</news:language></news:publication><news:publication_date>2026-10-20T17:20:00+09:00</news:publication_date><news:title>금융 · 코딩 · 법률 · 세금 일을 묶은 Vals Index v2.2 — 1위 Claude Sonnet 5.5 70.85%, 다만 '대체 모델' 답을 빼면 Gemini 4 Argon이 1위</news:title></news:news></url>
<url><loc>https://news.contentslab.ai/2026/10/20/aa-healthbench-length/</loc><news:news><news:publication><news:name>CONTENTSLAB NEWS</news:name><news:language>ko</news:language></news:publication><news:publication_date>2026-10-20T13:30:00+09:00</news:publication_date><news:title>의료 질문에 길게 답하면 점수가 깎여요 — HealthBench Professional 원점수 1위 Claude Opus 5.5, 길이 보정 뒤 6위</news:title></news:news></url>
<url><loc>https://news.contentslab.ai/2026/10/20/aa-personalitybench/</loc><news:news><news:publication><news:name>CONTENTSLAB NEWS</news:name><news:language>ko</news:language></news:publication><news:publication_date>2026-10-20T11:30:00+09:00</news:publication_date><news:title>AI 모델의 '말투'를 재는 평가가 나왔어요 — Artificial Analysis, 성격 특성 30개 · 지표 8개로 37개 모델 비교</news:title></news:news></url>
<url><loc>https://news.contentslab.ai/2026/10/20/openai-low-resource-languages/</loc><news:news><news:publication><news:name>CONTENTSLAB NEWS</news:name><news:language>ko</news:language></news:publication><news:publication_date>2026-10-20T08:30:00+09:00</news:publication_date><news:title>OpenAI, COBOL · Fortran 같은 옛 언어 코딩 성적 공개 — GPT-6 Astra는 20개 언어 평균 87.4%, COBOL 옮기기는 GPT-6.1 Sol이 97.9%</news:title></news:news></url>
<url><loc>https://news.contentslab.ai/2026/10/20/flux-3-max-video/</loc><news:news><news:publication><news:name>CONTENTSLAB NEWS</news:name><news:language>ko</news:language></news:publication><news:publication_date>2026-10-20T05:20:00+09:00</news:publication_date><news:title>10초 영상을 약 11~12초에 — fal·BFL 'FLUX 3 Max', 지금은 720p 초당 $0.0425 할인가</news:title></news:news></url>
<url><loc>https://news.contentslab.ai/2026/10/20/inception-mercury-3/</loc><news:news><news:publication><news:name>CONTENTSLAB NEWS</news:name><news:language>ko</news:language></news:publication><news:publication_date>2026-10-20T04:25:00+09:00</news:publication_date><news:title>Inception Mercury 3 공개 — 초당 1,068토큰, 자체 표 12개 중 8개에서 GPT-6 Luna · Claude Haiku 5.5보다 높아요</news:title></news:news></url>
<url><loc>https://news.contentslab.ai/2026/10/20/minimax-m31-flash/</loc><news:news><news:publication><news:name>CONTENTSLAB NEWS</news:name><news:language>ko</news:language></news:publication><news:publication_date>2026-10-20T03:20:00+09:00</news:publication_date><news:title>MiniMax M3.1 Flash 공개 — 입력 100만 토큰당 $0.15, 정체 숨긴 'Space Bunny'가 이 모델이었어요</news:title></news:news></url>
<url><loc>https://news.contentslab.ai/2026/10/20/exa-3/</loc><news:news><news:publication><news:name>CONTENTSLAB NEWS</news:name><news:language>ko</news:language></news:publication><news:publication_date>2026-10-20T02:17:00+09:00</news:publication_date><news:title>Exa 3.0, 검색 벤치마크 8개 모두 1위라고 발표… 1,000번에 $4 'Lite'도 새로 나와</news:title></news:news></url>
</urlset>
