{"id":1300,"date":"2026-04-03T11:08:41","date_gmt":"2026-04-03T11:08:41","guid":{"rendered":"https:\/\/bbotech.vn\/google-launches-gemma-4-apache-2-0-full-multimodal-and-the-open-model-race-intensifies\/"},"modified":"2026-04-03T11:11:23","modified_gmt":"2026-04-03T11:11:23","slug":"google-launches-gemma-4-apache-2-0-full-multimodal-and-the-open-model-race-intensifies","status":"publish","type":"post","link":"https:\/\/bbotech.vn\/vi\/google-launches-gemma-4-apache-2-0-full-multimodal-and-the-open-model-race-intensifies\/","title":{"rendered":"Google Ra M\u1eaft Gemma 4: Apache 2.0, Multimodal To\u00e0n B\u1ed9, v\u00e0 Cu\u1ed9c \u0110ua Open Model Ng\u00e0y C\u00e0ng N\u00f3ng"},"content":{"rendered":"<h2>B\u1ee9c tranh to\u00e0n c\u1ea3nh: Gemma 4 ra m\u1eaft trong b\u1ed1i c\u1ea3nh cu\u1ed9c \u0111ua m\u00f4 h\u00ecnh m\u1edf \u0111ang n\u00f3ng nh\u1ea5t t\u1eeb tr\u01b0\u1edbc \u0111\u1ebfn nay<\/h2>\n<p>Ng\u00e0y 2 th\u00e1ng 4 n\u0103m 2026, Google DeepMind ch\u00ednh th\u1ee9c ph\u00e1t h\u00e0nh <strong>Gemma 4<\/strong> \u2014 th\u1ebf h\u1ec7 m\u00f4 h\u00ecnh ng\u00f4n ng\u1eef m\u1edf m\u1edbi nh\u1ea5t c\u1ee7a h\u1ecd, v\u1edbi m\u1ed9t \u0111i\u1ec3m thay \u0111\u1ed5i \u0111\u00e1ng ch\u00fa \u00fd ngay t\u1eeb ng\u00e0y \u0111\u1ea7u: gi\u1ea5y ph\u00e9p Apache 2.0. Kh\u00f4ng c\u00f2n gi\u1ea5y ph\u00e9p t\u00f9y ch\u1ec9nh v\u1edbi \u0111i\u1ec1u kho\u1ea3n h\u1ea1n ch\u1ebf nh\u01b0 c\u00e1c phi\u00ean b\u1ea3n tr\u01b0\u1edbc \u2014 Gemma 4 \u0111\u01b0\u1ee3c ph\u00e9p d\u00f9ng th\u01b0\u01a1ng m\u1ea1i t\u1ef1 do, kh\u00f4ng r\u00e0ng bu\u1ed9c.<\/p>\n<p>\u0110\u00e2y kh\u00f4ng ch\u1ec9 l\u00e0 m\u1ed9t b\u1ea3n c\u1eadp nh\u1eadt k\u1ef9 thu\u1eadt. Trong b\u1ed1i c\u1ea3nh c\u00e1c m\u00f4 h\u00ecnh m\u1edf t\u1eeb Trung Qu\u1ed1c \u2014 Qwen, GLM, Kimi \u2014 \u0111ang \u0111\u1eb7t \u00e1p l\u1ef1c li\u00ean t\u1ee5c l\u00ean c\u00e1c lab ph\u01b0\u01a1ng T\u00e2y, Gemma 4 l\u00e0 c\u00e2u tr\u1ea3 l\u1eddi c\u1ee7a Google v\u1ec1 hi\u1ec7u n\u0103ng tr\u00ean t\u1eebng tham s\u1ed1, multimodal out-of-the-box, v\u00e0 kh\u1ea3 n\u0103ng ch\u1ea1y tr\u1ef1c ti\u1ebfp tr\u00ean thi\u1ebft b\u1ecb di \u0111\u1ed9ng.<\/p>\n<p>&nbsp;<\/p>\n<p><img loading=\"lazy\" decoding=\"async\" class=\"alignnone wp-image-1301 size-full\" src=\"https:\/\/bbotech.vn\/wp-content\/uploads\/2026\/04\/gemma-4.0.webp\" alt=\"Google Launches Gemma 4\" width=\"1200\" height=\"675\" srcset=\"https:\/\/bbotech.vn\/wp-content\/uploads\/2026\/04\/gemma-4.0.webp 1200w, https:\/\/bbotech.vn\/wp-content\/uploads\/2026\/04\/gemma-4.0-300x169.webp 300w, https:\/\/bbotech.vn\/wp-content\/uploads\/2026\/04\/gemma-4.0-1024x576.webp 1024w, https:\/\/bbotech.vn\/wp-content\/uploads\/2026\/04\/gemma-4.0-768x432.webp 768w\" sizes=\"auto, (max-width: 1200px) 100vw, 1200px\" \/><\/p>\n<p>&nbsp;<\/p>\n<p><!-- IMAGE: Google DeepMind Gemma 4 open source language model launch --><\/p>\n<h2>B\u1ed1n bi\u1ebfn th\u1ec3 \u2014 t\u1eeb mobile \u0111\u1ebfn server-grade reasoning<\/h2>\n<p>Gemma 4 ra m\u1eaft v\u1edbi <strong>b\u1ed1n phi\u00ean b\u1ea3n<\/strong> \u0111\u01b0\u1ee3c thi\u1ebft k\u1ebf cho c\u00e1c use case kh\u00e1c nhau:<\/p>\n<ul>\n<li><strong>Gemma 4 E2B (Effective 2B):<\/strong> T\u1ed1i \u01b0u cho thi\u1ebft b\u1ecb di \u0111\u1ed9ng, h\u1ed7 tr\u1ee3 \u1ea3nh, video, v\u00e0 \u00e2m thanh (nh\u1eadn d\u1ea1ng gi\u1ecdng n\u00f3i). Context window 128K token.<\/li>\n<li><strong>Gemma 4 E4B (Effective 4B):<\/strong> Phi\u00ean b\u1ea3n edge l\u1edbn h\u01a1n v\u1edbi c\u00f9ng b\u1ed9 multimodal input. M\u1ee5c ti\u00eau l\u00e0 on-device AI v\u1edbi hi\u1ec7u n\u0103ng cao h\u01a1n E2B trong \u0111i\u1ec1u ki\u1ec7n t\u00e0i nguy\u00ean h\u1ea1n ch\u1ebf.<\/li>\n<li><strong>Gemma 4 26B (Mixture-of-Experts):<\/strong> Ch\u1ec9 s\u1eed d\u1ee5ng <strong>3.8B tham s\u1ed1 active<\/strong> trong m\u1ed7i l\u1ea7n suy lu\u1eadn \u2014 MoE architecture gi\u00fap \u0111\u1ea1t reasoning m\u1ea1nh v\u1edbi chi ph\u00ed inference th\u1ea5p h\u01a1n nhi\u1ec1u so v\u1edbi dense model c\u00f9ng size. Context window 256K token.<\/li>\n<li><strong>Gemma 4 31B (Dense):<\/strong> M\u00f4 h\u00ecnh l\u1edbn nh\u1ea5t trong gia \u0111\u00ecnh, x\u1ebfp <strong>h\u1ea1ng 3 trong s\u1ed1 t\u1ea5t c\u1ea3 m\u00f4 h\u00ecnh m\u1edf<\/strong> tr\u00ean b\u1ea3ng x\u1ebfp h\u1ea1ng Arena AI text leaderboard to\u00e0n c\u1ea7u. Context window 256K token.<\/li>\n<\/ul>\n<p>T\u1ea5t c\u1ea3 b\u1ed1n phi\u00ean b\u1ea3n \u0111\u1ec1u <strong>x\u1eed l\u00fd \u1ea3nh v\u00e0 video natively<\/strong> \u2014 kh\u00f4ng ph\u1ea3i th\u00f4ng qua module ri\u00eang bi\u1ec7t. \u0110\u00e2y l\u00e0 \u0111i\u1ec3m kh\u00e1c bi\u1ec7t so v\u1edbi Gemma 3, v\u1ed1n ch\u1ec9 h\u1ed7 tr\u1ee3 h\u00ecnh \u1ea3nh \u1edf m\u1ed9t s\u1ed1 phi\u00ean b\u1ea3n nh\u1ea5t \u0111\u1ecbnh.<\/p>\n<p><!-- IMAGE: multimodal AI model architecture diagram image video audio --><\/p>\n<h2>Benchmark: M\u1ea1nh \u1edf reasoning, th\u1ef1c d\u1ee5ng \u1edf efficiency<\/h2>\n<p>C\u00e1c con s\u1ed1 benchmark c\u1ee7a Gemma 4 \u0111\u00e1ng ch\u00fa \u00fd, \u0111\u1eb7c bi\u1ec7t khi x\u00e9t theo t\u1ef7 l\u1ec7 hi\u1ec7u n\u0103ng \/ k\u00edch th\u01b0\u1edbc m\u00f4 h\u00ecnh:<\/p>\n<ul>\n<li><strong>Gemma 4 31B:<\/strong> MMLU Pro <strong>85.2%<\/strong>, AIME 2026 <strong>89.2%<\/strong>, Codeforces ELO <strong>2,150<\/strong>, LiveCodeBench v6 <strong>80.0%<\/strong><\/li>\n<li><strong>Gemma 4 26B MoE:<\/strong> AIME 2026 <strong>88.3%<\/strong>, GPQA Diamond <strong>82.3%<\/strong>, LiveCodeBench <strong>77.1%<\/strong> \u2014 v\u1edbi ch\u1ec9 3.8B active parameters<\/li>\n<\/ul>\n<p>Con s\u1ed1 89.2% tr\u00ean AIME 2026 (b\u00e0i thi to\u00e1n Olympic M\u1ef9) c\u1ee7a m\u00f4 h\u00ecnh 31B l\u00e0 \u0111\u1eb7c bi\u1ec7t \u1ea5n t\u01b0\u1ee3ng. V\u00e0 vi\u1ec7c 26B MoE \u0111\u1ea1t 88.3% v\u1edbi ch\u1ec9 3.8B active params khi\u1ebfn n\u00f3 tr\u1edf th\u00e0nh m\u1ed9t trong nh\u1eefng m\u00f4 h\u00ecnh reasoning hi\u1ec7u qu\u1ea3 nh\u1ea5t v\u1ec1 m\u1eb7t t\u00ednh to\u00e1n hi\u1ec7n c\u00f3.<\/p>\n<blockquote><p>Gemma 4 26B MoE \u0111\u1ea1t 88.3% tr\u00ean AIME 2026 v\u1edbi ch\u1ec9 3.8B tham s\u1ed1 active \u2014 m\u1ed9t trong nh\u1eefng t\u1ef7 l\u1ec7 hi\u1ec7u n\u0103ng\/chi ph\u00ed t\u1ed1t nh\u1ea5t trong ph\u00e2n kh\u00fac open model hi\u1ec7n nay.<\/p><\/blockquote>\n<p>Tuy nhi\u00ean, c\u1ea7n nh\u00ecn th\u1eb3ng v\u00e0o th\u1ef1c t\u1ebf: so v\u1edbi c\u00e1c m\u00f4 h\u00ecnh m\u1edf h\u00e0ng \u0111\u1ea7u t\u1eeb Trung Qu\u1ed1c \u2014 Qwen 3.5, GLM-5, Kimi K2.5 \u2014 Gemma 4 v\u1eabn \u0111\u1ee9ng sau m\u1ed9t kho\u1ea3ng, theo d\u1eef li\u1ec7u benchmark hi\u1ec7n c\u00f3. \u0110\u00e2y l\u00e0 \u00e1p l\u1ef1c m\u00e0 Google ph\u1ea3i ti\u1ebfp t\u1ee5c \u0111\u1ed1i m\u1eb7t trong cu\u1ed9c \u0111ua open model. <em>(Ngu\u1ed3n: Arena AI Leaderboard, th\u00e1ng 4\/2026)<\/em><\/p>\n<h2>Thay \u0111\u1ed5i l\u1edbn nh\u1ea5t kh\u00f4ng ph\u1ea3i k\u1ef9 thu\u1eadt: Apache 2.0 v\u00e0 \u00fd ngh\u0129a th\u1ef1c t\u1ebf<\/h2>\n<p>V\u1edbi c\u00e1c phi\u00ean b\u1ea3n Gemma tr\u01b0\u1edbc, Google d\u00f9ng gi\u1ea5y ph\u00e9p t\u00f9y ch\u1ec9nh \u2014 cho ph\u00e9p d\u00f9ng mi\u1ec5n ph\u00ed nh\u01b0ng k\u00e8m \u0111i\u1ec1u kho\u1ea3n h\u1ea1n ch\u1ebf \u1edf m\u1ed9t s\u1ed1 t\u00ecnh hu\u1ed1ng th\u01b0\u01a1ng m\u1ea1i. Gemma 4 chuy\u1ec3n ho\u00e0n to\u00e0n sang <strong>Apache 2.0<\/strong>.<\/p>\n<p>\u0110i\u1ec1u n\u00e0y c\u00f3 ngh\u0129a g\u00ec v\u1edbi engineering team?<\/p>\n<ul>\n<li><strong>Fine-tune v\u00e0 deploy th\u01b0\u01a1ng m\u1ea1i kh\u00f4ng h\u1ea1n ch\u1ebf<\/strong> \u2014 kh\u00f4ng c\u1ea7n lo ng\u1ea1i v\u1ec1 \u0111i\u1ec1u kho\u1ea3n gi\u1ea5y ph\u00e9p khi t\u00edch h\u1ee3p v\u00e0o s\u1ea3n ph\u1ea9m<\/li>\n<li><strong>Build derivative models<\/strong> d\u1ef1a tr\u00ean Gemma 4 v\u00e0 ph\u00e2n ph\u1ed1i t\u1ef1 do<\/li>\n<li><strong>T\u00edch h\u1ee3p v\u00e0o enterprise stack<\/strong> m\u00e0 kh\u00f4ng c\u1ea7n legal review ph\u1ee9c t\u1ea1p nh\u01b0 v\u1edbi gi\u1ea5y ph\u00e9p t\u00f9y ch\u1ec9nh<\/li>\n<\/ul>\n<p>\u0110\u00e2y l\u00e0 b\u01b0\u1edbc \u0111i th\u1ef1c d\u1ee5ng c\u1ee7a Google \u0111\u1ec3 c\u1ea1nh tranh tr\u1ef1c ti\u1ebfp v\u1edbi Meta&#8217;s Llama 3 \u2014 v\u1ed1n \u0111\u00e3 thu h\u00fat l\u01b0\u1ee3ng l\u1edbn developer nh\u1edd gi\u1ea5y ph\u00e9p permissive. <em>(Ngu\u1ed3n: Google Developers Blog, 2\/4\/2026)<\/em><\/p>\n<h2>Kh\u1ea3 n\u0103ng agentic v\u00e0 on-device \u2014 h\u01b0\u1edbng \u0111i cho n\u0103m 2026<\/h2>\n<p>M\u1ed9t trong nh\u1eefng \u0111i\u1ec3m nh\u1ea5n k\u1ef9 thu\u1eadt c\u1ee7a Gemma 4 l\u00e0 <strong>native support cho agentic tasks<\/strong>:<\/p>\n<ul>\n<li>Function calling \u0111\u01b0\u1ee3c t\u00edch h\u1ee3p s\u1eb5n \u2014 kh\u00f4ng c\u1ea7n wrapper b\u00ean ngo\u00e0i<\/li>\n<li>Structured JSON output h\u1ed7 tr\u1ee3 t\u00edch h\u1ee3p v\u1edbi API v\u00e0 tool pipeline<\/li>\n<li>System instruction handling nh\u1ea5t qu\u00e1n, quan tr\u1ecdng v\u1edbi multi-turn agent workflow<\/li>\n<\/ul>\n<p>V\u1edbi E2B v\u00e0 E4B ch\u1ea1y \u0111\u01b0\u1ee3c tr\u00ean mobile v\u1edbi context window 128K token v\u00e0 multimodal input bao g\u1ed3m c\u1ea3 audio, \u0111\u00e2y l\u00e0 n\u1ec1n t\u1ea3ng th\u1ef1c s\u1ef1 cho on-device AI agent \u2014 kh\u00f4ng c\u1ea7n g\u1ecdi v\u1ec1 server m\u1ed7i l\u1ea7n. Google \u0111ang h\u01b0\u1edbng t\u1edbi scenario m\u00e0 AI agent ch\u1ea1y ngay tr\u00ean \u0111i\u1ec7n tho\u1ea1i ng\u01b0\u1eddi d\u00f9ng, x\u1eed l\u00fd \u1ea3nh, gi\u1ecdng n\u00f3i, v\u00e0 video m\u00e0 kh\u00f4ng c\u1ea7n k\u1ebft n\u1ed1i cloud li\u00ean t\u1ee5c.<\/p>\n<p>M\u00f4 h\u00ecnh c\u00f3 th\u1ec3 t\u1ea3i ngay h\u00f4m nay t\u1eeb <strong>Hugging Face, Kaggle, v\u00e0 Ollama<\/strong> \u2014 kh\u00f4ng y\u00eau c\u1ea7u \u0111\u0103ng k\u00fd hay approval. <em>(Ngu\u1ed3n: Google DeepMind, 2\/4\/2026)<\/em><\/p>\n<h2>\u0110\u1ec1 xu\u1ea5t h\u00e0nh \u0111\u1ed9ng cho engineering v\u00e0 QA team<\/h2>\n<p>N\u1ebfu b\u1ea1n \u0111ang c\u00e2n nh\u1eafc t\u00edch h\u1ee3p LLM v\u00e0o s\u1ea3n ph\u1ea9m trong Q2-Q3 2026, \u0111\u00e2y l\u00e0 c\u00e1ch \u0111\u00e1nh gi\u00e1 Gemma 4 ph\u00f9 h\u1ee3p v\u1edbi context c\u1ee7a m\u00ecnh:<\/p>\n<ul>\n<li><strong>N\u1ebfu c\u1ea7n on-device:<\/strong> E2B v\u00e0 E4B l\u00e0 l\u1ef1a ch\u1ecdn th\u1ef1c t\u1ebf nh\u1ea5t hi\u1ec7n nay v\u1edbi multimodal + audio. Test tr\u1ef1c ti\u1ebfp tr\u00ean target device, \u0111\u1eebng ch\u1ec9 d\u1ef1a v\u00e0o benchmark.<\/li>\n<li><strong>N\u1ebfu c\u1ea7n reasoning n\u1eb7ng tr\u00ean server:<\/strong> 26B MoE cho t\u1ef7 l\u1ec7 hi\u1ec7u n\u0103ng\/chi ph\u00ed t\u1ed1t; 31B Dense n\u1ebfu c\u1ea7n top performance. So s\u00e1nh v\u1edbi Qwen 3.5 tr\u00ean c\u00f9ng benchmark domain tr\u01b0\u1edbc khi quy\u1ebft \u0111\u1ecbnh.<\/li>\n<li><strong>N\u1ebfu \u0111ang build agent:<\/strong> Native function calling c\u1ee7a Gemma 4 ti\u1ebft ki\u1ec7m engineering effort \u0111\u00e1ng k\u1ec3 so v\u1edbi prompt engineering th\u1ee7 c\u00f4ng. Nh\u01b0ng agent behavior c\u1ea7n \u0111\u01b0\u1ee3c ki\u1ec3m th\u1eed k\u1ef9 \u2014 function call accuracy kh\u00f4ng ph\u1ea3i l\u00fac n\u00e0o c\u0169ng t\u01b0\u01a1ng quan v\u1edbi benchmark score.<\/li>\n<\/ul>\n<p>Quan tr\u1ecdng: Apache 2.0 lo\u1ea1i b\u1ecf r\u1ee7i ro ph\u00e1p l\u00fd \u2014 nh\u01b0ng kh\u00f4ng lo\u1ea1i b\u1ecf r\u1ee7i ro ch\u1ea5t l\u01b0\u1ee3ng. Model output v\u1eabn c\u1ea7n validation pipeline, hallucination testing, v\u00e0 regression test khi b\u1ea1n fine-tune. \u0110\u00e2y l\u00e0 ph\u1ea7n m\u00e0 nhi\u1ec1u team b\u1ecf qua khi excited v\u1edbi m\u00f4 h\u00ecnh m\u1edbi.<\/p>\n<h2>K\u1ebft lu\u1eadn \u2014 g\u00f3c nh\u00ecn c\u1ee7a ch\u00fang t\u00f4i<\/h2>\n<p>Gemma 4 kh\u00f4ng ph\u1ea3i l\u00e0 m\u00f4 h\u00ecnh m\u1ea1nh nh\u1ea5t tr\u00ean th\u1ecb tr\u01b0\u1eddng t\u1ea1i th\u1eddi \u0111i\u1ec3m ra m\u1eaft. Nh\u01b0ng v\u1edbi Apache 2.0, multimodal out-of-the-box tr\u00ean t\u1ea5t c\u1ea3 phi\u00ean b\u1ea3n, v\u00e0 kh\u1ea3 n\u0103ng ch\u1ea1y tr\u00ean edge device \u2014 n\u00f3 c\u00f3 th\u1ec3 l\u00e0 m\u00f4 h\u00ecnh m\u1edf <em>th\u1ef1c d\u1ee5ng nh\u1ea5t<\/em> cho engineering team \u0111ang x\u00e2y d\u1ef1ng s\u1ea3n ph\u1ea9m th\u1ef1c t\u1ebf trong n\u0103m 2026. Cu\u1ed9c \u0111ua open model \u0111ang ng\u00e0y c\u00e0ng c\u00f3 l\u1ee3i cho ng\u01b0\u1eddi d\u00f9ng cu\u1ed1i \u2014 v\u00e0 \u0111i\u1ec1u \u0111\u00f3 s\u1ebd ti\u1ebfp t\u1ee5c t\u0103ng t\u1ed1c.<\/p>\n<p>\ud83d\udcf9 <strong>Xem video gi\u1edbi thi\u1ec7u ch\u00ednh th\u1ee9c t\u1eeb Google:<\/strong> <a href=\"https:\/\/www.youtube.com\/watch?v=jZVBoFOJK-Q\" target=\"_blank\" rel=\"noopener\">Gemma 4 \u2014 Google DeepMind (YouTube)<\/a><\/p>\n","protected":false},"excerpt":{"rendered":"<p>B\u1ee9c tranh to\u00e0n c\u1ea3nh: Gemma 4 ra m\u1eaft trong b\u1ed1i c\u1ea3nh cu\u1ed9c \u0111ua m\u00f4 h\u00ecnh m\u1edf \u0111ang n\u00f3ng nh\u1ea5t t\u1eeb tr\u01b0\u1edbc \u0111\u1ebfn nay Ng\u00e0y 2 th\u00e1ng 4 n\u0103m 2026, Google DeepMind ch\u00ednh th\u1ee9c ph\u00e1t h\u00e0nh Gemma 4 \u2014 th\u1ebf h\u1ec7 m\u00f4 h\u00ecnh ng\u00f4n ng\u1eef m\u1edf m\u1edbi nh\u1ea5t c\u1ee7a h\u1ecd, v\u1edbi m\u1ed9t \u0111i\u1ec3m thay \u0111\u1ed5i \u0111\u00e1ng [&hellip;]<\/p>\n","protected":false},"author":3,"featured_media":1301,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"footnotes":""},"categories":[25],"tags":[35,43,44],"class_list":["post-1300","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-tin-cong-nghe","tag-agentic-ai","tag-google","tag-google-launches-gemma-4"],"acf":[],"_links":{"self":[{"href":"https:\/\/bbotech.vn\/vi\/wp-json\/wp\/v2\/posts\/1300","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/bbotech.vn\/vi\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/bbotech.vn\/vi\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/bbotech.vn\/vi\/wp-json\/wp\/v2\/users\/3"}],"replies":[{"embeddable":true,"href":"https:\/\/bbotech.vn\/vi\/wp-json\/wp\/v2\/comments?post=1300"}],"version-history":[{"count":2,"href":"https:\/\/bbotech.vn\/vi\/wp-json\/wp\/v2\/posts\/1300\/revisions"}],"predecessor-version":[{"id":1303,"href":"https:\/\/bbotech.vn\/vi\/wp-json\/wp\/v2\/posts\/1300\/revisions\/1303"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/bbotech.vn\/vi\/wp-json\/wp\/v2\/media\/1301"}],"wp:attachment":[{"href":"https:\/\/bbotech.vn\/vi\/wp-json\/wp\/v2\/media?parent=1300"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/bbotech.vn\/vi\/wp-json\/wp\/v2\/categories?post=1300"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/bbotech.vn\/vi\/wp-json\/wp\/v2\/tags?post=1300"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}