{"id":13257,"date":"2026-11-17T10:45:11","date_gmt":"2026-11-17T18:45:11","guid":{"rendered":"https:\/\/devblogs.microsoft.com\/cosmosdb\/?p=13257"},"modified":"2026-11-16T16:54:56","modified_gmt":"2026-11-17T00:54:56","slug":"microsoft-embedding-one","status":"publish","type":"post","link":"https:\/\/devblogs.microsoft.com\/cosmosdb\/microsoft-embedding-one\/","title":{"rendered":"Simplify AI retrieval with embedding generation using Microsoft-Embedding-1\u00a0"},"content":{"rendered":"<p><span data-contrast=\"none\">Most enterprise knowledge doesn\u2019t live in clean text. It lives in charts, tables, scanned pages, and diagrams, where text-only search often misses it:<\/span><span data-ccp-props=\"{&quot;335559738&quot;:240,&quot;335559739&quot;:240}\">\u00a0<\/span><\/p>\n<ul>\n<li aria-setsize=\"-1\" data-leveltext=\"\uf0b7\" data-font=\"Symbol\" data-listid=\"2\" data-list-defn-props=\"{&quot;335552541&quot;:1,&quot;335559683&quot;:0,&quot;335559684&quot;:-2,&quot;335559685&quot;:720,&quot;335559991&quot;:360,&quot;469769226&quot;:&quot;Symbol&quot;,&quot;469769242&quot;:[8226],&quot;469777803&quot;:&quot;left&quot;,&quot;469777804&quot;:&quot;\uf0b7&quot;,&quot;469777815&quot;:&quot;hybridMultilevel&quot;}\" data-aria-posinset=\"1\" data-aria-level=\"1\"><span data-contrast=\"none\"><strong>Financial reports:<\/strong>\u00a0An analyst asks, \u201cWhich region\u2019s revenue fell in Q3?\u201d The answer is a bar in a chart on page 14, and no sentence on the page says it.<\/span><span data-ccp-props=\"{&quot;335559739&quot;:0}\">\u00a0<\/span><\/li>\n<\/ul>\n<ul>\n<li aria-setsize=\"-1\" data-leveltext=\"\uf0b7\" data-font=\"Symbol\" data-listid=\"2\" data-list-defn-props=\"{&quot;335552541&quot;:1,&quot;335559683&quot;:0,&quot;335559684&quot;:-2,&quot;335559685&quot;:720,&quot;335559991&quot;:360,&quot;469769226&quot;:&quot;Symbol&quot;,&quot;469769242&quot;:[8226],&quot;469777803&quot;:&quot;left&quot;,&quot;469777804&quot;:&quot;\uf0b7&quot;,&quot;469777815&quot;:&quot;hybridMultilevel&quot;}\" data-aria-posinset=\"2\" data-aria-level=\"1\"><span data-contrast=\"none\"><strong>Technical manuals:<\/strong> A field technician asks, \u201cWhat torque does the intake valve bolt need?\u201d The spec is a callout on an exploded-view diagram in a scanned PDF.<\/span><span data-ccp-props=\"{}\">\u00a0<\/span><\/li>\n<\/ul>\n<ul>\n<li aria-setsize=\"-1\" data-leveltext=\"\uf0b7\" data-font=\"Symbol\" data-listid=\"2\" data-list-defn-props=\"{&quot;335552541&quot;:1,&quot;335559683&quot;:0,&quot;335559684&quot;:-2,&quot;335559685&quot;:720,&quot;335559991&quot;:360,&quot;469769226&quot;:&quot;Symbol&quot;,&quot;469769242&quot;:[8226],&quot;469777803&quot;:&quot;left&quot;,&quot;469777804&quot;:&quot;\uf0b7&quot;,&quot;469777815&quot;:&quot;hybridMultilevel&quot;}\" data-aria-posinset=\"3\" data-aria-level=\"1\"><span data-contrast=\"none\"><strong>Labor reports:<\/strong> An analyst asks, \u201cWhich EU nationalities together make up almost 30% of all posted workers?\u201d The answer is the sum of a few slices of a pie chart.<\/span><span data-ccp-props=\"{}\">\u00a0<\/span><\/li>\n<\/ul>\n<ul>\n<li aria-setsize=\"-1\" data-leveltext=\"\uf0b7\" data-font=\"Symbol\" data-listid=\"2\" data-list-defn-props=\"{&quot;335552541&quot;:1,&quot;335559683&quot;:0,&quot;335559684&quot;:-2,&quot;335559685&quot;:720,&quot;335559991&quot;:360,&quot;469769226&quot;:&quot;Symbol&quot;,&quot;469769242&quot;:[8226],&quot;469777803&quot;:&quot;left&quot;,&quot;469777804&quot;:&quot;\uf0b7&quot;,&quot;469777815&quot;:&quot;hybridMultilevel&quot;}\" data-aria-posinset=\"5\" data-aria-level=\"1\"><span data-contrast=\"none\"><strong>Product catalogs:<\/strong> A shopper searches for \u201csomething to wear on a rainy hike.\u201d The listing says \u201clightweight waterproof jacket,\u201d and the photo shows a hooded shell on a muddy trail.<\/span><span data-ccp-props=\"{}\">\u00a0<\/span><\/li>\n<\/ul>\n<h2>Introducing Microsoft Embedding 1<\/h2>\n<p><span data-contrast=\"none\">Today we&#8217;re introducing <\/span><b><span data-contrast=\"none\">Microsoft Embedding 1<\/span><\/b><span data-contrast=\"none\">, a state-of-the-art multimodal embedding model developed by <\/span>Azure Cosmos DB and Microsoft Research<span data-contrast=\"none\">.\u00a0It embeds text, images, and full document pages into one vector space, so a single text query can find the passage, photo, or chart that answers it. The model is available through <\/span><a href=\"https:\/\/aka.ms\/CosmosEmbeddingGenerationSetup\"><b><span data-contrast=\"none\">Azure Data AI<\/span><\/b><\/a><span data-contrast=\"none\"> today and is coming soon to <\/span>Microsoft Foundry.<\/p>\n<p style=\"text-align: center;\"><iframe loading=\"lazy\" src=\"\/\/www.youtube.com\/embed\/TrN71ZKgSJM\" width=\"660\" height=\"370\" allowfullscreen=\"allowfullscreen\"><\/iframe><\/p>\n<h2 aria-level=\"2\"><b><span data-contrast=\"none\">State-of-the-art retrieval on visual documents<\/span><\/b><span data-ccp-props=\"{&quot;335559738&quot;:299,&quot;335559739&quot;:299}\">\u00a0<\/span><\/h2>\n<p><a href=\"https:\/\/mteb-leaderboard.hf.space\/benchmark\/ViDoRe(v3)\"><span data-contrast=\"none\">ViDoRe v3<\/span><\/a><span data-contrast=\"none\"> part of the MTEB leaderboard, is a benchmark for retrieval over visually rich enterprise documents such as reports and manuals. On ViDoRe v3, Microsoft Embedding 1 ranks <strong>#1<\/strong> in its size class with a score of 66.24 nDCG@10, as of November 16, 2026.<\/span><\/p>\n<p><span class=\"TextRun SCXW130412933 BCX8\" lang=\"EN-US\" xml:lang=\"EN-US\" data-contrast=\"none\"><span class=\"NormalTextRun SCXW130412933 BCX8\">Microsoft Embedding 1 (microsoft-embedding-1-large) is trained <\/span><span class=\"NormalTextRun SCXW130412933 BCX8\">to output b<\/span><span class=\"NormalTextRun SCXW130412933 BCX8\">oth <\/span><span class=\"NormalTextRun SCXW130412933 BCX8\">a dense retrieval<\/span><span class=\"NormalTextRun SCXW130412933 BCX8\"> model and a late-interaction model. You can request one dense vector per input, a multi-vector representation, or both. Both representations score ahead of Cohere Embed v5 Pro, Gemma Embedding 2, voyage-multimodal-3.5, and oth<\/span><span class=\"NormalTextRun SpellingErrorV2Themed SCXW130412933 BCX8\">ers on<\/span> <span class=\"NormalTextRun SpellingErrorV2Themed SCXW130412933 BCX8\">ViDoRe<\/span><span class=\"NormalTextRun SCXW130412933 BCX8\"> v3.\u00a0<\/span><\/span><\/p>\n<p><span class=\"TextRun SCXW28819839 BCX8\" lang=\"EN-US\" xml:lang=\"EN-US\" data-contrast=\"none\"><span class=\"NormalTextRun SCXW28819839 BCX8\">Late-interaction models are increasingly used for precise retrieval over complex content, where text-only dense retrieval struggles most. They emit one vector per input token and score a pair of inputs (text, image, or document page) with MaxSim over their vectors. On <\/span><span class=\"NormalTextRun SpellingErrorV2Themed SCXW28819839 BCX8\">ViDoRe<\/span><span class=\"NormalTextRun SCXW28819839 BCX8\"> v3, multi-vector <\/span><span class=\"NormalTextRun SCXW28819839 BCX8\">representations<\/span> <span class=\"NormalTextRun ContextualSpellingAndGrammarErrorV2Themed SCXW28819839 BCX8\">add<\/span> <span class=\"NormalTextRun SCXW28819839 BCX8\">nearly 6<\/span><span class=\"NormalTextRun SCXW28819839 BCX8\"> points of nDCG@10 over dense output.\u00a0<\/span><\/span><span class=\"EOP Selected SCXW28819839 BCX8\" data-ccp-props=\"{&quot;134233117&quot;:false,&quot;134233118&quot;:false,&quot;335551550&quot;:0,&quot;335551620&quot;:0,&quot;335559738&quot;:240,&quot;335559739&quot;:240}\">\u00a0<\/span><\/p>\n<p><span class=\"TextRun SCXW66520194 BCX8\" lang=\"EN-US\" xml:lang=\"EN-US\" data-contrast=\"none\"><span class=\"NormalTextRun SCXW66520194 BCX8\">Storing one vector per token can be costly, though. For a page with about 1,250 visual tokens, a late-interaction model can output 1,250 <\/span><span class=\"NormalTextRun ContextualSpellingAndGrammarErrorV2Themed SCXW66520194 BCX8\">vectors<\/span><span class=\"NormalTextRun SCXW66520194 BCX8\"> \u00d7 1,024 dimensions, which is about 2.5 MB. To address this, we trained the model so you can control the output size by limiting:\u00a0<\/span><\/span><span class=\"EOP Selected SCXW66520194 BCX8\" data-ccp-props=\"{&quot;134233117&quot;:false,&quot;134233118&quot;:false,&quot;335551550&quot;:0,&quot;335551620&quot;:0,&quot;335559738&quot;:240,&quot;335559739&quot;:240}\">\u00a0<\/span><\/p>\n<ol>\n<li aria-setsize=\"-1\" data-leveltext=\"%1.\" data-font=\"Aptos\" data-listid=\"7\" data-list-defn-props=\"{&quot;335551671&quot;:1,&quot;335552541&quot;:0,&quot;335559683&quot;:0,&quot;335559684&quot;:-1,&quot;335559685&quot;:720,&quot;335559991&quot;:360,&quot;469769242&quot;:[65533,0,46],&quot;469777803&quot;:&quot;left&quot;,&quot;469777804&quot;:&quot;%1.&quot;,&quot;469777815&quot;:&quot;hybridMultilevel&quot;}\" data-aria-posinset=\"1\" data-aria-level=\"1\"><span data-contrast=\"none\">The number of vectors per input.<\/span><span data-ccp-props=\"{&quot;134233117&quot;:false,&quot;134233118&quot;:false,&quot;335551550&quot;:0,&quot;335551620&quot;:0,&quot;335559738&quot;:0,&quot;335559739&quot;:0}\">\u00a0<\/span><\/li>\n<li aria-setsize=\"-1\" data-leveltext=\"%1.\" data-font=\"Aptos\" data-listid=\"7\" data-list-defn-props=\"{&quot;335551671&quot;:1,&quot;335552541&quot;:0,&quot;335559683&quot;:0,&quot;335559684&quot;:-1,&quot;335559685&quot;:720,&quot;335559991&quot;:360,&quot;469769242&quot;:[65533,0,46],&quot;469777803&quot;:&quot;left&quot;,&quot;469777804&quot;:&quot;%1.&quot;,&quot;469777815&quot;:&quot;hybridMultilevel&quot;}\" data-aria-posinset=\"1\" data-aria-level=\"1\"><span data-contrast=\"none\">The dimensions per vector.<\/span><span data-ccp-props=\"{&quot;134233117&quot;:false,&quot;134233118&quot;:false,&quot;335551550&quot;:0,&quot;335551620&quot;:0,&quot;335559738&quot;:0,&quot;335559739&quot;:0}\">\u00a0<\/span><\/li>\n<\/ol>\n<p><span data-contrast=\"none\">For example, you can request 32 vectors with 128 dimensions each in float16, which comes to 8 KB per page. That\u2019s in the same ballpark as a single dense vector. The model still maintains strong relevance, giving up about 3.4 points of nDCG@10 for a roughly 300-fold reduction in storage.<\/span><span data-ccp-props=\"{&quot;134233117&quot;:false,&quot;134233118&quot;:false,&quot;335551550&quot;:0,&quot;335551620&quot;:0,&quot;335559738&quot;:240,&quot;335559739&quot;:240}\">\u00a0<\/span><\/p>\n<h3 aria-level=\"3\"><b><span data-contrast=\"none\">Text retrieval<\/span><\/b><span data-ccp-props=\"{&quot;134233117&quot;:false,&quot;134233118&quot;:false,&quot;335551550&quot;:0,&quot;335551620&quot;:0,&quot;335559738&quot;:281,&quot;335559739&quot;:281}\">\u00a0<\/span><\/h3>\n<p><span data-contrast=\"none\">On text retrieval, <a href=\"http:\/\/aka.ms\/MicrosoftEmbeddingOne\">Microsoft Embedding 1<\/a> <span class=\"NormalTextRun SCXW102163537 BCX8\">scores<\/span><span class=\"NormalTextRun SCXW102163537 BCX8\">\u00a0on-par with<\/span><span class=\"NormalTextRun SCXW102163537 BCX8\"> OpenAI text-embedding-3-large on the BEIR benchmark (15 tasks, nDCG@10), with vectors about 17% smaller (2,560 vs. 3,072 dimensions). At 1,536 dimensions, the same size as text-embedding-3-<\/span><span class=\"NormalTextRun ContextualSpellingAndGrammarErrorV2Themed SCXW102163537 BCX8\">small,<\/span><span class=\"NormalTextRun SCXW102163537 BCX8\"> it scores 3.8 points higher.<\/span><span class=\"NormalTextRun SCXW102163537 BCX8\">\u00a0<\/span><\/span><\/p>\n<table style=\"border-collapse: collapse; width: 100%;\">\n<tbody>\n<tr>\n<td style=\"width: 33.3333%; text-align: center;\"><b><span data-contrast=\"none\">Model<\/span><\/b><span data-ccp-props=\"{&quot;134233117&quot;:false,&quot;134233118&quot;:false,&quot;335551550&quot;:2,&quot;335551620&quot;:2,&quot;335559738&quot;:0,&quot;335559739&quot;:0}\">\u00a0<\/span><\/td>\n<td style=\"width: 22.2458%; text-align: center;\" data-celllook=\"0\"><b><span data-contrast=\"none\">Dimensions<\/span><\/b><span data-ccp-props=\"{&quot;134233117&quot;:false,&quot;134233118&quot;:false,&quot;335551550&quot;:2,&quot;335551620&quot;:2,&quot;335559738&quot;:0,&quot;335559739&quot;:0}\">\u00a0<\/span><\/td>\n<td style=\"width: 25.2119%; text-align: center;\" data-celllook=\"0\"><b><span data-contrast=\"none\">BEIR average<\/span><\/b><span data-ccp-props=\"{&quot;134233117&quot;:false,&quot;134233118&quot;:false,&quot;335551550&quot;:2,&quot;335551620&quot;:2,&quot;335559738&quot;:0,&quot;335559739&quot;:0}\">\u00a0<\/span><\/td>\n<\/tr>\n<tr aria-rowindex=\"2\">\n<td style=\"width: 33.3333%; text-align: center;\" data-celllook=\"0\"><span data-contrast=\"none\">Microsoft Embedding 1<\/span><span data-ccp-props=\"{&quot;134233117&quot;:false,&quot;134233118&quot;:false,&quot;335551550&quot;:0,&quot;335551620&quot;:0,&quot;335559738&quot;:0,&quot;335559739&quot;:0}\">\u00a0<\/span><\/td>\n<td style=\"width: 22.2458%; text-align: center;\" data-celllook=\"0\"><span data-contrast=\"none\">1,536<\/span><span data-ccp-props=\"{&quot;134233117&quot;:false,&quot;134233118&quot;:false,&quot;335551550&quot;:0,&quot;335551620&quot;:0,&quot;335559738&quot;:0,&quot;335559739&quot;:0}\">\u00a0<\/span><\/td>\n<td style=\"width: 25.2119%; text-align: center;\" data-celllook=\"0\"><span data-contrast=\"none\">54.92<\/span><span data-ccp-props=\"{&quot;134233117&quot;:false,&quot;134233118&quot;:false,&quot;335551550&quot;:0,&quot;335551620&quot;:0,&quot;335559738&quot;:0,&quot;335559739&quot;:0}\">\u00a0<\/span><\/td>\n<\/tr>\n<tr aria-rowindex=\"3\">\n<td style=\"width: 33.3333%; text-align: center;\" data-celllook=\"0\"><span data-contrast=\"none\">Microsoft Embedding 1<\/span><span data-ccp-props=\"{&quot;134233117&quot;:false,&quot;134233118&quot;:false,&quot;335551550&quot;:0,&quot;335551620&quot;:0,&quot;335559738&quot;:0,&quot;335559739&quot;:0}\">\u00a0<\/span><\/td>\n<td style=\"width: 22.2458%; text-align: center;\" data-celllook=\"0\"><span data-contrast=\"none\">2,560<\/span><span data-ccp-props=\"{&quot;134233117&quot;:false,&quot;134233118&quot;:false,&quot;335551550&quot;:0,&quot;335551620&quot;:0,&quot;335559738&quot;:0,&quot;335559739&quot;:0}\">\u00a0<\/span><\/td>\n<td style=\"width: 25.2119%; text-align: center;\" data-celllook=\"0\"><span data-contrast=\"none\">55.29<\/span><span data-ccp-props=\"{&quot;134233117&quot;:false,&quot;134233118&quot;:false,&quot;335551550&quot;:0,&quot;335551620&quot;:0,&quot;335559738&quot;:0,&quot;335559739&quot;:0}\">\u00a0<\/span><\/td>\n<\/tr>\n<tr aria-rowindex=\"4\">\n<td style=\"width: 33.3333%; text-align: center;\" data-celllook=\"0\"><span data-contrast=\"none\">OpenAI text-embedding-3-small<\/span><span data-ccp-props=\"{&quot;134233117&quot;:false,&quot;134233118&quot;:false,&quot;335551550&quot;:0,&quot;335551620&quot;:0,&quot;335559738&quot;:0,&quot;335559739&quot;:0}\">\u00a0<\/span><\/td>\n<td style=\"width: 22.2458%; text-align: center;\" data-celllook=\"0\"><span data-contrast=\"none\">1,536<\/span><span data-ccp-props=\"{&quot;134233117&quot;:false,&quot;134233118&quot;:false,&quot;335551550&quot;:0,&quot;335551620&quot;:0,&quot;335559738&quot;:0,&quot;335559739&quot;:0}\">\u00a0<\/span><\/td>\n<td style=\"width: 25.2119%; text-align: center;\" data-celllook=\"0\"><span data-contrast=\"none\">51.08<\/span><span data-ccp-props=\"{&quot;134233117&quot;:false,&quot;134233118&quot;:false,&quot;335551550&quot;:0,&quot;335551620&quot;:0,&quot;335559738&quot;:0,&quot;335559739&quot;:0}\">\u00a0<\/span><\/td>\n<\/tr>\n<tr aria-rowindex=\"5\">\n<td style=\"width: 33.3333%; text-align: center;\" data-celllook=\"0\"><span data-contrast=\"none\">OpenAI text-embedding-3-large<\/span><span data-ccp-props=\"{&quot;134233117&quot;:false,&quot;134233118&quot;:false,&quot;335551550&quot;:0,&quot;335551620&quot;:0,&quot;335559738&quot;:0,&quot;335559739&quot;:0}\">\u00a0<\/span><\/td>\n<td style=\"width: 22.2458%; text-align: center;\" data-celllook=\"0\"><span data-contrast=\"none\">3,072<\/span><span data-ccp-props=\"{&quot;134233117&quot;:false,&quot;134233118&quot;:false,&quot;335551550&quot;:0,&quot;335551620&quot;:0,&quot;335559738&quot;:0,&quot;335559739&quot;:0}\">\u00a0<\/span><\/td>\n<td style=\"width: 25.2119%; text-align: center;\" data-celllook=\"0\"><span data-contrast=\"none\">55.43<\/span><span data-ccp-props=\"{&quot;134233117&quot;:false,&quot;134233118&quot;:false,&quot;335551550&quot;:0,&quot;335551620&quot;:0,&quot;335559738&quot;:0,&quot;335559739&quot;:0}\">\u00a0<\/span><\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p><span class=\"TextRun SCXW189094717 BCX8\" lang=\"EN-US\" xml:lang=\"EN-US\" data-contrast=\"none\"><span class=\"NormalTextRun SCXW189094717 BCX8\">Microsoft Embedding 1\u2019s largest gains over text-embedding-3-large are on <\/span><span class=\"NormalTextRun SpellingErrorV2Themed SCXW189094717 BCX8\">ClimateFEVER<\/span><span class=\"NormalTextRun SCXW189094717 BCX8\"> (+6.3), <\/span><span class=\"NormalTextRun SpellingErrorV2Themed SCXW189094717 BCX8\">HotpotQA<\/span><span class=\"NormalTextRun SCXW189094717 BCX8\"> (+3.6), Natural Questions (+3.3), <\/span><span class=\"NormalTextRun SpellingErrorV2Themed SCXW189094717 BCX8\">DBPedia<\/span><span class=\"NormalTextRun SCXW189094717 BCX8\"> (+3.0), and MS MARCO (+1.8). text-embedding-3-large <\/span><span class=\"NormalTextRun ContextualSpellingAndGrammarErrorV2Themed SCXW189094717 BCX8\">leads most on<\/span><span class=\"NormalTextRun SCXW189094717 BCX8\"> TREC-COVID. Microsoft Embedding 1 scores are measured with fp16 vectors, as served.\u00a0<\/span><\/span><span class=\"EOP Selected SCXW189094717 BCX8\" data-ccp-props=\"{&quot;134233117&quot;:false,&quot;134233118&quot;:false,&quot;335551550&quot;:0,&quot;335551620&quot;:0,&quot;335559738&quot;:240,&quot;335559739&quot;:240}\">\u00a0<\/span><\/p>\n<h3 aria-level=\"3\"><b><span data-contrast=\"none\">Choose your balance of accuracy and storage<\/span><\/b><span data-ccp-props=\"{&quot;134233117&quot;:false,&quot;134233118&quot;:false,&quot;335551550&quot;:0,&quot;335551620&quot;:0,&quot;335559738&quot;:281,&quot;335559739&quot;:281}\">\u00a0<\/span><\/h3>\n<p><span data-contrast=\"none\">Two settings let you trade accuracy against storage:\u00a0<\/span><span data-ccp-props=\"{&quot;134233117&quot;:false,&quot;134233118&quot;:false,&quot;335551550&quot;:0,&quot;335551620&quot;:0,&quot;335559738&quot;:240,&quot;335559739&quot;:240}\">\u00a0<\/span><\/p>\n<ul>\n<li aria-setsize=\"-1\" data-leveltext=\"\uf0b7\" data-font=\"Symbol\" data-listid=\"12\" data-list-defn-props=\"{&quot;335552541&quot;:1,&quot;335559683&quot;:0,&quot;335559684&quot;:-2,&quot;335559685&quot;:720,&quot;335559991&quot;:360,&quot;469769226&quot;:&quot;Symbol&quot;,&quot;469769242&quot;:[8226],&quot;469777803&quot;:&quot;left&quot;,&quot;469777804&quot;:&quot;\uf0b7&quot;,&quot;469777815&quot;:&quot;hybridMultilevel&quot;}\" data-aria-posinset=\"1\" data-aria-level=\"1\"><b><span data-contrast=\"none\">Multi-vector output.<\/span><\/b><span data-contrast=\"none\"> Choose how many vectors the model returns per page or passage. A reasonable starting point is 32 vectors per input with 128 dimensions each. On ViDoRe v3, even this setting beats the best dense result by 2.5 points.\u00a0<\/span><span data-ccp-props=\"{&quot;134233117&quot;:false,&quot;134233118&quot;:false,&quot;335551550&quot;:0,&quot;335551620&quot;:0,&quot;335559738&quot;:0,&quot;335559739&quot;:0}\">\u00a0<\/span><\/li>\n<\/ul>\n<ul>\n<li aria-setsize=\"-1\" data-leveltext=\"\uf0b7\" data-font=\"Symbol\" data-listid=\"12\" data-list-defn-props=\"{&quot;335552541&quot;:1,&quot;335559683&quot;:0,&quot;335559684&quot;:-2,&quot;335559685&quot;:720,&quot;335559991&quot;:360,&quot;469769226&quot;:&quot;Symbol&quot;,&quot;469769242&quot;:[8226],&quot;469777803&quot;:&quot;left&quot;,&quot;469777804&quot;:&quot;\uf0b7&quot;,&quot;469777815&quot;:&quot;hybridMultilevel&quot;}\" data-aria-posinset=\"2\" data-aria-level=\"1\"><b><span data-contrast=\"none\">Flexible dimensions.<\/span><\/b><span data-contrast=\"none\"> With Matryoshka representation learning (MRL), you can pick a dense vector size of 1,024, 1,536, 2,048, or 2,560 dimensions. Smaller vectors cost less to store and are slightly faster to search.\u00a0<\/span><span data-ccp-props=\"{&quot;134233117&quot;:false,&quot;134233118&quot;:false,&quot;335551550&quot;:0,&quot;335551620&quot;:0,&quot;335559738&quot;:0,&quot;335559739&quot;:0}\">\u00a0<\/span><\/li>\n<\/ul>\n<p style=\"text-align: center;\"><i><span data-contrast=\"none\"><a href=\"https:\/\/devblogs.microsoft.com\/cosmosdb\/wp-content\/uploads\/sites\/52\/2026\/11\/videorev3.png\"><img decoding=\"async\" class=\"alignnone wp-image-13264 size-full\" src=\"https:\/\/devblogs.microsoft.com\/cosmosdb\/wp-content\/uploads\/sites\/52\/2026\/11\/videorev3.png\" alt=\"Line chart of ViDoRe v3 retrieval quality (nDCG@10) versus average storage per document on a log scale. Dense scores 57.08 to 60.37 at 2 to 5 KB. Multi-vector capped at 32 vectors scores 62.88 to 64.18 at 8 to 64 KB. Unconstrained multi-vector scores 65.63 to 66.24 at about 300 to 2,500 KB.\" width=\"1600\" height=\"868\" srcset=\"https:\/\/devblogs.microsoft.com\/cosmosdb\/wp-content\/uploads\/sites\/52\/2026\/11\/videorev3.png 1600w, https:\/\/devblogs.microsoft.com\/cosmosdb\/wp-content\/uploads\/sites\/52\/2026\/11\/videorev3-300x163.png 300w, https:\/\/devblogs.microsoft.com\/cosmosdb\/wp-content\/uploads\/sites\/52\/2026\/11\/videorev3-1024x556.png 1024w, https:\/\/devblogs.microsoft.com\/cosmosdb\/wp-content\/uploads\/sites\/52\/2026\/11\/videorev3-768x417.png 768w, https:\/\/devblogs.microsoft.com\/cosmosdb\/wp-content\/uploads\/sites\/52\/2026\/11\/videorev3-1536x833.png 1536w\" sizes=\"(max-width: 1600px) 100vw, 1600px\" \/><\/a>\nFigure: ViDoRe v3 retrieval quality (nDCG@10) versus average raw FP16 vector storage per document. Each point is a different vector dimension.\n<\/span><\/i><\/p>\n<p><span data-contrast=\"none\">The chart above shows how each configuration trades retrieval quality for storage on ViDoRe v3. Capping multi-vector output at 32 vectors keeps most of the quality gain over dense vectors at a fraction of the storage of unconstrained multi-vector output.\u00a0<\/span><span data-ccp-props=\"{&quot;134233117&quot;:false,&quot;134233118&quot;:false,&quot;335551550&quot;:0,&quot;335551620&quot;:0,&quot;335559738&quot;:240,&quot;335559739&quot;:240}\">\u00a0<\/span><\/p>\n<p><span data-contrast=\"none\">For text-heavy content, dense vectors work well. For visual documents such as reports and manuals, multi-vector output gives the highest retrieval quality. Multi-vector output requires a retrieval system that supports MaxSim scoring.\u00a0<\/span><span data-ccp-props=\"{&quot;134233117&quot;:false,&quot;134233118&quot;:false,&quot;335551550&quot;:0,&quot;335551620&quot;:0,&quot;335559738&quot;:240,&quot;335559739&quot;:240}\">\u00a0<\/span><\/p>\n<p><b><span data-contrast=\"none\">Generate embeddings in three steps<\/span><\/b><span data-ccp-props=\"{&quot;134233117&quot;:false,&quot;134233118&quot;:false,&quot;335551550&quot;:0,&quot;335551620&quot;:0,&quot;335559738&quot;:281,&quot;335559739&quot;:281}\">\u00a0<\/span><\/p>\n<p><span data-contrast=\"none\">You can use Microsoft Embedding 1 without deploying or hosting a model:<\/span><span data-ccp-props=\"{&quot;134233117&quot;:false,&quot;134233118&quot;:false,&quot;335551550&quot;:0,&quot;335551620&quot;:0,&quot;335559738&quot;:240,&quot;335559739&quot;:240}\">\u00a0<\/span><\/p>\n<ol>\n<li aria-setsize=\"-1\" data-leveltext=\"%1.\" data-font=\"Aptos\" data-listid=\"9\" data-list-defn-props=\"{&quot;335551671&quot;:1,&quot;335552541&quot;:0,&quot;335559683&quot;:0,&quot;335559684&quot;:-1,&quot;335559685&quot;:720,&quot;335559991&quot;:360,&quot;469769242&quot;:[65533,0,46],&quot;469777803&quot;:&quot;left&quot;,&quot;469777804&quot;:&quot;%1.&quot;,&quot;469777815&quot;:&quot;hybridMultilevel&quot;}\" data-aria-posinset=\"1\" data-aria-level=\"1\"><b><span data-contrast=\"none\">Enable embedding generation.<\/span><\/b><span data-contrast=\"none\"> Today, you enable it on an Azure Cosmos DB account, and inference is provisioned for you.<\/span><\/li>\n<li aria-setsize=\"-1\" data-leveltext=\"%1.\" data-font=\"Aptos\" data-listid=\"9\" data-list-defn-props=\"{&quot;335551671&quot;:1,&quot;335552541&quot;:0,&quot;335559683&quot;:0,&quot;335559684&quot;:-1,&quot;335559685&quot;:720,&quot;335559991&quot;:360,&quot;469769242&quot;:[65533,0,46],&quot;469777803&quot;:&quot;left&quot;,&quot;469777804&quot;:&quot;%1.&quot;,&quot;469777815&quot;:&quot;hybridMultilevel&quot;}\" data-aria-posinset=\"1\" data-aria-level=\"1\"><b><span data-contrast=\"none\">Generate embeddings<\/span><\/b><span data-contrast=\"none\"> by calling the endpoint with the Azure Data AI client library (<\/span><span data-contrast=\"none\">azure-data-ai<\/span><span data-contrast=\"none\">) for Python or .NET.<\/span><span data-ccp-props=\"{&quot;134233117&quot;:false,&quot;134233118&quot;:false,&quot;335551550&quot;:0,&quot;335551620&quot;:0,&quot;335559738&quot;:0,&quot;335559739&quot;:0}\">\u00a0<\/span><\/li>\n<li aria-setsize=\"-1\" data-leveltext=\"%1.\" data-font=\"Aptos\" data-listid=\"9\" data-list-defn-props=\"{&quot;335551671&quot;:1,&quot;335552541&quot;:0,&quot;335559683&quot;:0,&quot;335559684&quot;:-1,&quot;335559685&quot;:720,&quot;335559991&quot;:360,&quot;469769242&quot;:[65533,0,46],&quot;469777803&quot;:&quot;left&quot;,&quot;469777804&quot;:&quot;%1.&quot;,&quot;469777815&quot;:&quot;hybridMultilevel&quot;}\" data-aria-posinset=\"1\" data-aria-level=\"1\"><b><span data-contrast=\"none\">Store and search the vectors<\/span><\/b><span data-contrast=\"none\"> in your vector store, such as Azure Cosmos DB for dense vectors.<\/span><span data-ccp-props=\"{&quot;134233117&quot;:false,&quot;134233118&quot;:false,&quot;335551550&quot;:0,&quot;335551620&quot;:0,&quot;335559738&quot;:0,&quot;335559739&quot;:0}\">\u00a0<\/span><\/li>\n<\/ol>\n<p style=\"text-align: center;\"><iframe loading=\"lazy\" src=\"\/\/www.youtube.com\/embed\/1uzGq4Ytw4M\" width=\"660\" height=\"370\" allowfullscreen=\"allowfullscreen\"><\/iframe><\/p>\n<div class=\"me1-code-tabs\"><\/div>\n<div class=\"me1-code-tabs\"><\/div>\n<div class=\"me1-code-tabs\"><input id=\"me1-tab-py\" checked=\"checked\" name=\"me1-lang\" type=\"radio\" \/> <label for=\"me1-tab-py\">Python<\/label> <input id=\"me1-tab-net\" name=\"me1-lang\" type=\"radio\" \/> <label for=\"me1-tab-net\">.NET<\/label><\/p>\n<div class=\"me1-panel me1-panel-py\">\n<p>Install the Azure Identity library and the preview of the Azure Data AI library:<\/p>\n<pre><code class=\"language-bash\">pip install azure-identity\r\npip install \"azure-data-ai==1.1.0b1\"<\/code><\/pre>\n<p>This example embeds a search question and a page from a scanned equipment manual. Set your endpoint, image URL, and SAS token as environment variables.<\/p>\n<pre><code class=\"language-python\">import os\r\n\r\nfrom azure.data.ai import InferenceClient\r\nfrom azure.identity import DefaultAzureCredential\r\n\r\nwith DefaultAzureCredential() as credential:\r\n    with InferenceClient(\r\n        endpoint=os.environ[\"EMBEDDING_ENDPOINT\"],\r\n        credential=credential,\r\n    ) as client:\r\n        response = client.get_embeddings(\r\n            model=\"microsoft-embedding-1-large\",\r\n            inputs=[\r\n                {\r\n                    \"kind\": \"query\",\r\n                    \"content\": \"What torque does the intake valve bolt need?\",\r\n                },\r\n                {\r\n                    \"kind\": \"image\",\r\n                    \"url\": os.environ[\"IMAGE_BLOB_URL\"],\r\n                    \"sasToken\": os.environ[\"IMAGE_SAS_TOKEN\"],\r\n                },\r\n            ],\r\n        )\r\n\r\n# Results are returned in input order.\r\nembeddings = [item.embedding for item in response.data]<\/code><\/pre>\n<\/div>\n<div class=\"me1-panel me1-panel-net\">\n<p>Install the preview of the Azure Data AI library and the Azure Identity library:<\/p>\n<pre><code class=\"language-bash\">dotnet add package Azure.Data.AI --prerelease\r\ndotnet add package Azure.Identity<\/code><\/pre>\n<p>This example embeds a search question and a page from a scanned equipment manual. Set your endpoint, image URL, and SAS token as environment variables.<\/p>\n<pre><code class=\"language-csharp\">using Azure;\r\nusing Azure.Data.AI;\r\nusing Azure.Identity;\r\n\r\nvar client = new InferenceClient(\r\n    new Uri(Environment.GetEnvironmentVariable(\"EMBEDDING_ENDPOINT\")!),\r\n    new DefaultAzureCredential());\r\n\r\nResponse&lt;EmbeddingResult&gt; response = client.GetEmbeddings(\r\n    model: \"microsoft-embedding-1-large\",\r\n    inputs: new[]\r\n    {\r\n        new EmbeddingsInputContent\r\n        {\r\n            Kind = EmbeddingContentKind.Query,\r\n            Content = \"What torque does the intake valve bolt need?\"\r\n        },\r\n        new EmbeddingsInputContent\r\n        {\r\n            Kind = EmbeddingContentKind.Image,\r\n            Uri = new Uri(Environment.GetEnvironmentVariable(\"IMAGE_BLOB_URL\")!),\r\n            SasToken = Environment.GetEnvironmentVariable(\"IMAGE_SAS_TOKEN\")\r\n        }\r\n    });\r\n\r\n\/\/ Results are returned in input order.\r\nvar embeddings = response.Value.Data\r\n    .Cast&lt;MultimodalDenseEmbedding&gt;()\r\n    .Select(item =&gt; item.Embedding.ToObjectFromJson&lt;float[]&gt;())\r\n    .ToArray();<\/code><\/pre>\n<\/div>\n<\/div>\n<style>\n  .me1-code-tabs { margin: 1.5em 0; }\n  .me1-code-tabs input[type=\"radio\"] {\n    position: absolute; opacity: 0; width: 1px; height: 1px; pointer-events: none;\n  }\n  .me1-code-tabs label {\n    display: inline-block !important; padding: 8px 18px !important; margin: 0 4px 0 0 !important;\n    cursor: pointer; font-weight: 600; color: inherit; opacity: 0.7;\n    border: none !important; border-bottom: 3px solid transparent !important;\n    background: none !important; box-shadow: none !important; outline: none;\n  }\n  .me1-code-tabs label:hover { opacity: 1; }\n  .me1-code-tabs input:checked + label {\n    opacity: 1; color: #4da3ff; border-bottom-color: #4da3ff !important;\n  }\n  .me1-code-tabs input:focus-visible + label { outline: 2px solid #4da3ff; outline-offset: 2px; }\n  .me1-code-tabs .me1-panel {\n    display: none; padding-top: 0.5em;\n  }\n  #me1-tab-py:checked ~ .me1-panel-py,\n  #me1-tab-net:checked ~ .me1-panel-net { display: block; }<\/p>\n<p>  \/* Keep indentation: the site's styles collapse leading spaces. *\/\n  .me1-code-tabs pre,\n  .me1-code-tabs pre code,\n  .me1-code-tabs pre code * {\n    white-space: pre !important;\n    tab-size: 4;\n  }\n  .me1-code-tabs pre { overflow-x: auto !important; }\n<\/style>\n<h3 aria-level=\"3\"><b><span data-contrast=\"none\">Turn embeddings into relevant results<\/span><\/b><span data-ccp-props=\"{&quot;134233117&quot;:false,&quot;134233118&quot;:false,&quot;335559738&quot;:281,&quot;335559739&quot;:281}\">\u00a0<\/span><\/h3>\n<p><span data-contrast=\"none\">Store each embedding with its content and metadata, embed the user\u2019s question with <\/span><span data-contrast=\"none\">kind: &#8220;query&#8221;<\/span><span data-contrast=\"none\">, and retrieve the closest items. In Azure Cosmos DB, store dense vectors in a container with a matching vector policy and index, and query them with `<\/span><span data-contrast=\"none\">VectorDistance`<\/span><span data-contrast=\"none\">. Show the results directly, or pass them to an AI model as grounding context.<\/span><span data-ccp-props=\"{&quot;134233117&quot;:false,&quot;134233118&quot;:false,&quot;335559738&quot;:240,&quot;335559739&quot;:240}\">\u00a0<\/span><\/p>\n<h2><span data-ccp-props=\"{}\">\u00a0<\/span><b><span data-contrast=\"none\">Pay for what you use<\/span><\/b><span data-ccp-props=\"{&quot;134245418&quot;:true,&quot;201341983&quot;:0,&quot;335559738&quot;:135,&quot;335559739&quot;:70,&quot;335559740&quot;:280}\">\u00a0<\/span><\/h2>\n<p><span data-contrast=\"none\">We charge for the input tokens you process, so you can start with a small dataset and expand as your application needs grow.<\/span><span data-ccp-props=\"{}\">\u00a0<\/span><\/p>\n<table style=\"border-collapse: collapse; width: 100%;\">\n<tbody>\n<tr aria-rowindex=\"1\">\n<td style=\"width: 17.9634%; text-align: center;\" data-celllook=\"69905\"><b><span data-contrast=\"none\">Input<\/span><\/b><span data-ccp-props=\"{&quot;201341983&quot;:0,&quot;335559739&quot;:0,&quot;335559740&quot;:230}\">\u00a0<\/span><\/td>\n<td style=\"width: 82.0366%; text-align: center;\" data-celllook=\"69905\"><b><span data-contrast=\"none\">Price per 1 million input tokens<\/span><\/b><span data-ccp-props=\"{&quot;201341983&quot;:0,&quot;335559739&quot;:0,&quot;335559740&quot;:230}\">\u00a0<\/span><\/td>\n<\/tr>\n<tr aria-rowindex=\"2\">\n<td style=\"width: 17.9634%; text-align: center;\" data-celllook=\"69905\"><span data-contrast=\"none\">Text<\/span><span data-ccp-props=\"{&quot;201341983&quot;:0,&quot;335559739&quot;:0,&quot;335559740&quot;:230}\">\u00a0<\/span><\/td>\n<td style=\"width: 82.0366%; text-align: center;\" data-celllook=\"69905\"><span data-contrast=\"none\">USD $0.12\u00a0<\/span><span data-ccp-props=\"{&quot;201341983&quot;:0,&quot;335559739&quot;:0,&quot;335559740&quot;:230}\">\u00a0<\/span><\/td>\n<\/tr>\n<tr aria-rowindex=\"3\">\n<td style=\"width: 17.9634%; text-align: center;\" data-celllook=\"69905\"><span data-contrast=\"none\">Images<\/span><span data-ccp-props=\"{&quot;201341983&quot;:0,&quot;335559739&quot;:0,&quot;335559740&quot;:230}\">\u00a0<\/span><\/td>\n<td style=\"width: 82.0366%; text-align: center;\" data-celllook=\"69905\"><span data-contrast=\"none\">USD $0.40\u00a0<\/span><span data-ccp-props=\"{&quot;201341983&quot;:0,&quot;335559739&quot;:0,&quot;335559740&quot;:230}\">\u00a0<\/span><\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<h2>Conclusion<\/h2>\n<p>Charts, tables, and scanned pages hold some of your most important answers, and text-only search can\u2019t see them. Microsoft Embedding 1 can, with the highest published ViDoRe v3 score for a single retrieval model and no model to deploy or host. Pick the balance of accuracy and storage that fits your app, and pay only for what you embed. Your best answers are already in your documents. Now your search can find them.<\/p>\n<p><span data-ccp-props=\"{}\"><div  class=\"d-flex justify-content-center\"><a class=\"cta_button_link btn-primary mb-24\" href=\"https:\/\/aka.ms\/MicrosoftEmbeddingOne\" target=\"_blank\">Get Started Today!<\/a><\/div>\u00a0<\/span><\/p>\n<h4>About Azure Cosmos DB<\/h4>\n<p>Azure Cosmos DB is a fully managed NoSQL and vector database for modern applications, including AI apps. With SLA-backed performance and availability, automatic scaling, and global distribution, it helps developers build fast, reliable applications.<\/p>\n<p>Find more developer resources at the <a id=\"menur5e2\" class=\"fui-Link ___1q1shib f2hkw1w f3rmtva f1ewtqcl fyind8e f1k6fduh f1w7gpdv fk6fouc fjoy568 figsok6 f1s184ao f1mk8lai fnbmjn9 f1o700av f13mvf36 f1cmlufx f9n3di6 f1ids18y f1tx3yz7 f1deo86v f1eh06m1 f1iescvh fhgqx19 f1olyrje f1p93eir f1nev41a f1h8hb77 f1lqvz6u f10aw75t fsle3fq f17ae5zn\" title=\"https:\/\/cosmosdb.microsoft.com\/\" href=\"https:\/\/cosmosdb.microsoft.com\" target=\"_blank\" rel=\"noreferrer noopener\" aria-label=\"Link Azure Cosmos DB developer hub\">Azure Cosmos DB developer hub<\/a>.<\/p>\n<p>Stay connected with Azure Cosmos DB on <a id=\"menur5e5\" class=\"fui-Link ___1q1shib f2hkw1w f3rmtva f1ewtqcl fyind8e f1k6fduh f1w7gpdv fk6fouc fjoy568 figsok6 f1s184ao f1mk8lai fnbmjn9 f1o700av f13mvf36 f1cmlufx f9n3di6 f1ids18y f1tx3yz7 f1deo86v f1eh06m1 f1iescvh fhgqx19 f1olyrje f1p93eir f1nev41a f1h8hb77 f1lqvz6u f10aw75t fsle3fq f17ae5zn\" title=\"https:\/\/twitter.com\/azurecosmosdb\" href=\"https:\/\/twitter.com\/AzureCosmosDB\" target=\"_blank\" rel=\"noreferrer noopener\" aria-label=\"Link X\">X<\/a>, <a id=\"menur5e8\" class=\"fui-Link ___1q1shib f2hkw1w f3rmtva f1ewtqcl fyind8e f1k6fduh f1w7gpdv fk6fouc fjoy568 figsok6 f1s184ao f1mk8lai fnbmjn9 f1o700av f13mvf36 f1cmlufx f9n3di6 f1ids18y f1tx3yz7 f1deo86v f1eh06m1 f1iescvh fhgqx19 f1olyrje f1p93eir f1nev41a f1h8hb77 f1lqvz6u f10aw75t fsle3fq f17ae5zn\" title=\"https:\/\/aka.ms\/azurecosmosdbyoutube\" href=\"https:\/\/aka.ms\/AzureCosmosDBYouTube\" target=\"_blank\" rel=\"noreferrer noopener\" aria-label=\"Link YouTube\">YouTube<\/a>, and <a id=\"menur5eb\" class=\"fui-Link ___1q1shib f2hkw1w f3rmtva f1ewtqcl fyind8e f1k6fduh f1w7gpdv fk6fouc fjoy568 figsok6 f1s184ao f1mk8lai fnbmjn9 f1o700av f13mvf36 f1cmlufx f9n3di6 f1ids18y f1tx3yz7 f1deo86v f1eh06m1 f1iescvh fhgqx19 f1olyrje f1p93eir f1nev41a f1h8hb77 f1lqvz6u f10aw75t fsle3fq f17ae5zn\" title=\"https:\/\/www.linkedin.com\/company\/azure-cosmos-db\/\" href=\"https:\/\/www.linkedin.com\/company\/azure-cosmos-db\/\" target=\"_blank\" rel=\"noreferrer noopener\" aria-label=\"Link LinkedIn\">LinkedIn<\/a>.<\/p>\n","protected":false},"excerpt":{"rendered":"<p>Most enterprise knowledge doesn\u2019t live in clean text. It lives in charts, tables, scanned pages, and diagrams, where text-only search often misses it:\u00a0 Financial reports:\u00a0An analyst asks, \u201cWhich region\u2019s revenue fell in Q3?\u201d The answer is a bar in a chart on page 14, and no sentence on the page says it.\u00a0 Technical manuals: A [&hellip;]<\/p>\n","protected":false},"author":125132,"featured_media":13274,"comment_status":"open","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"footnotes":""},"categories":[1610],"tags":[],"class_list":["post-13257","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-ai"],"acf":[],"blog_post_summary":"<p>Most enterprise knowledge doesn\u2019t live in clean text. It lives in charts, tables, scanned pages, and diagrams, where text-only search often misses it:\u00a0 Financial reports:\u00a0An analyst asks, \u201cWhich region\u2019s revenue fell in Q3?\u201d The answer is a bar in a chart on page 14, and no sentence on the page says it.\u00a0 Technical manuals: A [&hellip;]<\/p>\n","_links":{"self":[{"href":"https:\/\/devblogs.microsoft.com\/cosmosdb\/wp-json\/wp\/v2\/posts\/13257","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/devblogs.microsoft.com\/cosmosdb\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/devblogs.microsoft.com\/cosmosdb\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/devblogs.microsoft.com\/cosmosdb\/wp-json\/wp\/v2\/users\/125132"}],"replies":[{"embeddable":true,"href":"https:\/\/devblogs.microsoft.com\/cosmosdb\/wp-json\/wp\/v2\/comments?post=13257"}],"version-history":[{"count":2,"href":"https:\/\/devblogs.microsoft.com\/cosmosdb\/wp-json\/wp\/v2\/posts\/13257\/revisions"}],"predecessor-version":[{"id":13316,"href":"https:\/\/devblogs.microsoft.com\/cosmosdb\/wp-json\/wp\/v2\/posts\/13257\/revisions\/13316"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/devblogs.microsoft.com\/cosmosdb\/wp-json\/wp\/v2\/media\/13274"}],"wp:attachment":[{"href":"https:\/\/devblogs.microsoft.com\/cosmosdb\/wp-json\/wp\/v2\/media?parent=13257"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/devblogs.microsoft.com\/cosmosdb\/wp-json\/wp\/v2\/categories?post=13257"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/devblogs.microsoft.com\/cosmosdb\/wp-json\/wp\/v2\/tags?post=13257"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}