{"id":1171497,"date":"2026-05-21T10:00:00","date_gmt":"2026-05-21T17:00:00","guid":{"rendered":""},"modified":"2026-05-21T16:28:24","modified_gmt":"2026-05-21T23:28:24","slug":"magenticlite-magenticbrain-fara1-5-an-agentic-experience-optimized-for-small-models","status":"publish","type":"post","link":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/blog\/magenticlite-magenticbrain-fara1-5-an-agentic-experience-optimized-for-small-models\/","title":{"rendered":"MagenticLite, MagenticBrain, Fara1.5: An agentic experience optimized for small models"},"content":{"rendered":"\n<figure class=\"wp-block-image size-full\"><img loading=\"lazy\" decoding=\"async\" width=\"2560\" height=\"1441\" src=\"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-content\/uploads\/2026\/05\/MagenticLite-BlogHeroFeature-1400x788-1-scaled.jpg\" alt=\"MagenticLite\" class=\"wp-image-1171504\" srcset=\"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-content\/uploads\/2026\/05\/MagenticLite-BlogHeroFeature-1400x788-1-scaled.jpg 2560w, https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-content\/uploads\/2026\/05\/MagenticLite-BlogHeroFeature-1400x788-1-300x169.jpg 300w, https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-content\/uploads\/2026\/05\/MagenticLite-BlogHeroFeature-1400x788-1-1024x576.jpg 1024w, https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-content\/uploads\/2026\/05\/MagenticLite-BlogHeroFeature-1400x788-1-768x432.jpg 768w, https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-content\/uploads\/2026\/05\/MagenticLite-BlogHeroFeature-1400x788-1-1536x865.jpg 1536w, https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-content\/uploads\/2026\/05\/MagenticLite-BlogHeroFeature-1400x788-1-2048x1153.jpg 2048w, https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-content\/uploads\/2026\/05\/MagenticLite-BlogHeroFeature-1400x788-1-1066x600.jpg 1066w, https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-content\/uploads\/2026\/05\/MagenticLite-BlogHeroFeature-1400x788-1-655x368.jpg 655w, https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-content\/uploads\/2026\/05\/MagenticLite-BlogHeroFeature-1400x788-1-240x135.jpg 240w, https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-content\/uploads\/2026\/05\/MagenticLite-BlogHeroFeature-1400x788-1-640x360.jpg 640w, https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-content\/uploads\/2026\/05\/MagenticLite-BlogHeroFeature-1400x788-1-960x540.jpg 960w, https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-content\/uploads\/2026\/05\/MagenticLite-BlogHeroFeature-1400x788-1-1280x720.jpg 1280w, https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-content\/uploads\/2026\/05\/MagenticLite-BlogHeroFeature-1400x788-1-1920x1080.jpg 1920w\" sizes=\"auto, (max-width: 2560px) 100vw, 2560px\" \/><\/figure>\n\n\n\n<div style=\"padding-bottom:0; padding-top:0\" class=\"wp-block-msr-immersive-section alignfull row wp-block-msr-immersive-section\">\n\t\n\t<div class=\"container\">\n\t\t<div class=\"wp-block-msr-immersive-section__inner wp-block-msr-immersive-section__inner--narrow\">\n\t\t\t<div class=\"wp-block-columns mb-10 pb-1 pr-1 is-layout-flex wp-container-core-columns-is-layout-9d6595d7 wp-block-columns-is-layout-flex\" style=\"box-shadow:var(--wp--preset--shadow--outlined)\">\n<div class=\"wp-block-column is-layout-flow wp-block-column-is-layout-flow\">\n<h2 class=\"wp-block-heading h3\" id=\"at-a-glance\">At a glance<\/h2>\n\n\n\n<ul class=\"wp-block-list\">\n<li>MagenticLite is an agentic application that works across both the browser and local file system in a single workflow. Built as the next generation of Magentic-UI, it combines a redesigned app with a harness optimized for small models.<\/li>\n\n\n\n<li>MagenticBrain and Fara1.5 are small models designed for orchestration and computer-use tasks, respectively. Fara1.5 is the next iteration of Fara and delivers measurable gains on real-world browser tasks.<\/li>\n\n\n\n<li>Together, these releases explore how far agentic performance can be pushed with smaller models, codesigned tools, and an optimized execution harness.<\/li>\n<\/ul>\n<\/div>\n<\/div>\t\t<\/div>\n\t<\/div>\n\n\t<\/div>\n\n\n\n<p>Today, <a href=\"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/lab\/ai-frontiers\/\">Microsoft Research AI Frontiers<\/a> releases <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"https:\/\/aka.ms\/MagenticLite\" type=\"link\" id=\"https:\/\/aka.ms\/MagenticLite\" target=\"_blank\" rel=\"noopener noreferrer\">MagenticLite<span class=\"sr-only\"> (opens in new tab)<\/span><\/a>, an experimental agentic application designed for small models. As the next generation of Magentic-UI, it works across the browser and local file system in a single workflow.<\/p>\n\n\n\n<p>MagenticLite is powered by two purpose-built models: MagenticBrain, for reasoning, delegation, and terminal use, and Fara1.5, a computer-use model family for browser-based tasks. The three components were designed to work together as a single system. The result is an agent that runs efficiently, keeps data on the user\u2019s machine, and supports a broad range of agentic tasks. It also points toward a broader goal: capable agents that can run directly on users\u2019 hardware.<\/p>\n\n\n\n<p>The project is built around a key research bet: that agentic capability depends on tool orchestration and action rather than knowledge alone. That insight makes it possible to use smaller models while still enabling a broad range of agentic tasks at a fraction of the cost.<\/p>\n\n\n\n<p>MagenticLite also reflects how we approach agentic AI end-to-end\u2014from training data and model design to orchestration, interaction design, and human oversight throughout the experience.<\/p>\n\n\n\n<figure class=\"wp-block-image aligncenter size-full\"><img loading=\"lazy\" decoding=\"async\" width=\"1560\" height=\"780\" src=\"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-content\/uploads\/2026\/05\/magentic_releases.png\" alt=\"Figure 1 \u2013 One experience, three components.png | A diagram titled \"3 Related Releases\" showing three side-by-side panels, each representing one component of a unified agentic system. The first panel (pink) is labeled \"Agentic Application\" and features MagenticLite, described as an agentic experience optimized for SLMs that works across Browser and File System. The second panel (purple) is labeled \"Orchestrator Model\" and features MagenticBrain, described as handling reasoning, coding, and delegation, and introduced as a new orchestrator model. The third panel (teal) is labeled \"Computer Use Model\" and features Fara1.5, described as handling web navigation and the browser, and achieving new state-of-the-art performance on web tasks in its weight class. Arrows from all three panels converge downward to the text \"One efficient agentic experience.\"\" class=\"wp-image-1172866\" srcset=\"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-content\/uploads\/2026\/05\/magentic_releases.png 1560w, https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-content\/uploads\/2026\/05\/magentic_releases-300x150.png 300w, https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-content\/uploads\/2026\/05\/magentic_releases-1024x512.png 1024w, https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-content\/uploads\/2026\/05\/magentic_releases-768x384.png 768w, https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-content\/uploads\/2026\/05\/magentic_releases-1536x768.png 1536w, https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-content\/uploads\/2026\/05\/magentic_releases-240x120.png 240w\" sizes=\"auto, (max-width: 1560px) 100vw, 1560px\" \/><figcaption class=\"wp-element-caption\">Figure 1. One experience, three components: MagenticLite, MagenticBrain, and Fara1.5. <\/figcaption><\/figure>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"included-in-this-release\">Included in this release<\/h2>\n\n\n\n<p id=\"magenticlite\"><a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"https:\/\/aka.ms\/MagenticLite\" target=\"_blank\" rel=\"noopener noreferrer\"><strong>MagenticLite<\/strong><span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<p>The next generation of <a href=\"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/blog\/magentic-ui-an-experimental-human-centered-web-agent\/\" type=\"link\" id=\"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/blog\/magentic-ui-an-experimental-human-centered-web-agent\/\">Magentic-UI<\/a>, our experimental agentic experience, is powered by an agent harness rebuilt for small models, with an updated user interface informed by community feedback. It works across users\u2019 browsers and local file systems in a single workflow.<\/p>\n\n\n\n<p id=\"magenticbrain\"><strong><a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" rel=\"noopener noreferrer\" target=\"_blank\" href=\"https:\/\/aka.ms\/MagenticBrain-foundry\">MagenticBrain<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/strong><\/p>\n\n\n\n<p>MagenticBrain&nbsp;is&nbsp;MagenticLite\u2019s&nbsp;planner, coder, and delegator in one.&nbsp;It&nbsp;turns&nbsp;vague&nbsp;requests&nbsp;into concrete plans,&nbsp;selects&nbsp;the right tool or subagent for each step,&nbsp;writes&nbsp;code when&nbsp;needed, and recovers&nbsp;should&nbsp;something break mid-task.&nbsp;<\/p>\n\n\n\n<p id=\"fara1-5\"><a href=\"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/articles\/fara1-5-computer-use-agent\/\" type=\"link\" id=\"https:\/\/aka.ms\/fara\" target=\"_blank\" rel=\"noreferrer noopener\"><strong>Fara1.5<\/strong><\/a><\/p>\n\n\n\n<p>The next generation of our computer-use model family,&nbsp;Fara1.5&nbsp;comes&nbsp;&nbsp;in&nbsp;three sizes,&nbsp;with a&nbsp;flagship 9-billion-parameter&nbsp;model&nbsp;for&nbsp;most use cases.&nbsp;Fara1.5 sets new&nbsp;state-of-the-art&nbsp;(SOTA)&nbsp;results among small computer-use models and nearly doubles Fara-7B&#8217;s performance on web navigation, with sharper handling of forms, credentialed sites, and long-running tasks.<\/p>\n\n\n\n<p>Each component is useful on its own, but they work best together. Codesigning the app, models, and the harness enables capable and reliable agentic performance at this scale.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"our-research-approach-doing-more-with-less\">Our research approach: Doing more with less<\/h3>\n\n\n\n<p>We started with a simple question: what does it take to make a small model genuinely good at agentic tasks? The answer spanned the full lifecycle\u2014data generation, training objectives, model design, and orchestration had to be redesigned together rather than in isolation.<\/p>\n\n\n\n<p>We identified requirements from real-world use cases like filling out forms, conducting browser research, and managing files locally, and built an evaluation dataset around them. Standard benchmarks capture part of the picture, but they are not always a direct measure of real-world usefulness. Scenario-based evaluations complemented those benchmarks and became a key signal for iterative improvement across both the models and the harness, as shown in Figure 2.<\/p>\n\n\n\n<figure class=\"wp-block-image aligncenter size-full\"><img loading=\"lazy\" decoding=\"async\" width=\"1560\" height=\"933\" src=\"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-content\/uploads\/2026\/05\/Figure-2-Eval-flywheel-e1778687653105.png\" alt=\"Figure 2 \u2013 Eval flywheel.png | A flowchart titled \"Eval-Driven Improvement Flywheel\" illustrating a three-stage iterative loop for building agentic systems. At the top, a box labeled \"Define Success\" describes it as the north star orienting every decision, encompassing Scenarios, Tasks, and Metrics. An arrow leads down to a central box labeled \"Evaluate & Diagnose,\" which involves evaluating each component and the stack as a whole, and diagnosing failures to attribute root causes. From this central box, bidirectional arrows connect to two parallel boxes at the bottom: \"Improve Harness\" on the left (optimizing the agentic layer wrapping models, covering Instructions, Context management, and Tool handling) and \"Train Models\" on the right (identifying and creating data to close key gaps, referencing MagenticBrain and Fara1.5). Dashed arrows from both bottom boxes feed back into the Evaluate & Diagnose stage, forming a continuous loop.\" class=\"wp-image-1171808\" srcset=\"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-content\/uploads\/2026\/05\/Figure-2-Eval-flywheel-e1778687653105.png 1560w, https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-content\/uploads\/2026\/05\/Figure-2-Eval-flywheel-e1778687653105-300x179.png 300w, https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-content\/uploads\/2026\/05\/Figure-2-Eval-flywheel-e1778687653105-1024x612.png 1024w, https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-content\/uploads\/2026\/05\/Figure-2-Eval-flywheel-e1778687653105-768x459.png 768w, https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-content\/uploads\/2026\/05\/Figure-2-Eval-flywheel-e1778687653105-1536x919.png 1536w, https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-content\/uploads\/2026\/05\/Figure-2-Eval-flywheel-e1778687653105-240x144.png 240w\" sizes=\"auto, (max-width: 1560px) 100vw, 1560px\" \/><figcaption class=\"wp-element-caption\">Figure 2. An iterative process for building agentic systems involves defining success criteria, evaluating performance, and refining the models or system design (or both). Then repeat.<\/figcaption><\/figure>\n\n\n\n<p>For the user experience, we retained key elements from Magentic-UI, including visibility into the agent&#8217;s reasoning and actions, the ability for users to take direct control, and explicit approval at critical points. Based on <a href=\"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-content\/uploads\/2025\/07\/magentic-ui-report.pdf?msockid=1d972c2941d266e30aae39e740b967e6\" type=\"link\" id=\"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-content\/uploads\/2025\/07\/magentic-ui-report.pdf?msockid=1d972c2941d266e30aae39e740b967e6\" target=\"_blank\" rel=\"noreferrer noopener\">recent user studies<\/a>, we also made MagenticLite easier to learn and collaborate with through updated browser and chat views, designed to make it easier for users to understand the agent\u2019s actions and intervene when needed. This is illustrated in Figure 3.<\/p>\n\n\n\n<figure class=\"wp-block-image aligncenter size-full\"><img loading=\"lazy\" decoding=\"async\" width=\"2496\" height=\"1425\" src=\"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-content\/uploads\/2026\/05\/Figure-3-MAGUI-new-interface-scaled-e1778687236569.png\" alt=\"Figure 3 \u2013 MAGUI new interface.png | A screenshot of the MagenticLite 2.0.063 application interface. The left sidebar shows a session history with task names and statuses, including one active task highlighted in pink. The central panel displays an ongoing agent session with a sequential log of actions\u2014including \"Used web browser,\" \"Memorized a fact,\" and another \"Used web browser\" entry\u2014along with a text input field at the bottom. The right panel shows a live Browser History view with a web search results page rendered inside it. A disclaimer at the bottom of the interface reads: \"MagenticLite can make mistakes. Please monitor its work and intervene if necessary.\"\" class=\"wp-image-1171810\" srcset=\"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-content\/uploads\/2026\/05\/Figure-3-MAGUI-new-interface-scaled-e1778687236569.png 2496w, https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-content\/uploads\/2026\/05\/Figure-3-MAGUI-new-interface-scaled-e1778687236569-300x171.png 300w, https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-content\/uploads\/2026\/05\/Figure-3-MAGUI-new-interface-scaled-e1778687236569-1024x585.png 1024w, https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-content\/uploads\/2026\/05\/Figure-3-MAGUI-new-interface-scaled-e1778687236569-768x438.png 768w, https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-content\/uploads\/2026\/05\/Figure-3-MAGUI-new-interface-scaled-e1778687236569-1536x877.png 1536w, https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-content\/uploads\/2026\/05\/Figure-3-MAGUI-new-interface-scaled-e1778687236569-2048x1169.png 2048w, https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-content\/uploads\/2026\/05\/Figure-3-MAGUI-new-interface-scaled-e1778687236569-240x137.png 240w\" sizes=\"auto, (max-width: 2496px) 100vw, 2496px\" \/><figcaption class=\"wp-element-caption\">Figure 3. MagenticLite\u2019s interface includes updated browser and chat views designed to make it easier to understand agent actions and intervene when needed.<\/figcaption><\/figure>\n\n\n\n\t<div class=\"border-bottom border-top border-gray-300 mt-5 mb-5 msr-promo text-center text-md-left alignwide\" data-bi-aN=\"promo\" data-bi-id=\"1141385\">\n\t\t\n\n\t\n\t<div class=\"row pt-3 pb-4 align-items-center\">\n\t\t\t\t\t\t<div class=\"msr-promo__media col-12 col-md-5\">\n\t\t\t\t<a class=\"bg-gray-300 display-block\" href=\"https:\/\/ai.azure.com\/labs\" aria-label=\"Azure AI Foundry Labs\" data-bi-cN=\"Azure AI Foundry Labs\" target=\"_blank\">\n\t\t\t\t\t<img decoding=\"async\" class=\"w-100 display-block\" src=\"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-content\/uploads\/2025\/06\/Azure-AI-Foundry_1600x900.jpg\" \/>\n\t\t\t\t<\/a>\n\t\t\t<\/div>\n\t\t\t\n\t\t\t<div class=\"msr-promo__content p-3 px-5 col-12 col-md\">\n\n\t\t\t\t\t\t\t\t\t<h2 class=\"h4\">Azure AI Foundry Labs<\/h2>\n\t\t\t\t\n\t\t\t\t\t\t\t\t<p id=\"azure-ai-foundry-labs\" class=\"large\">Get a glimpse of potential future directions for AI, with these experimental technologies from Microsoft Research.<\/p>\n\t\t\t\t\n\t\t\t\t\t\t\t\t<div class=\"wp-block-buttons justify-content-center justify-content-md-start\">\n\t\t\t\t\t<div class=\"wp-block-button\">\n\t\t\t\t\t\t<a href=\"https:\/\/ai.azure.com\/labs\" aria-describedby=\"azure-ai-foundry-labs\" class=\"btn btn-brand glyph-append glyph-append-chevron-right\" data-bi-cN=\"Azure AI Foundry Labs\" target=\"_blank\">\n\t\t\t\t\t\t\tAzure AI Foundry\t\t\t\t\t\t<\/a>\n\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t\t\t\t<\/div><!--\/.msr-promo__content-->\n\t<\/div><!--\/.msr-promo__inner-wrap-->\n\t<\/div><!--\/.msr-promo-->\n\t\n\n\n<h2 class=\"wp-block-heading\" id=\"system-components\">System components<\/h2>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"fara1-5-a-computer-use-model-that-outperforms-its-weight-class\">Fara1.5: A&nbsp;computer-use&nbsp;model&nbsp;that&nbsp;outperforms&nbsp;its&nbsp;weight&nbsp;class<\/h3>\n\n\n\n<p>Fara1.5&nbsp;is the next generation of our computer-use model family,&nbsp;which is available&nbsp;in three sizes,&nbsp;with a&nbsp;flagship&nbsp;9B&nbsp;model&nbsp;recommended&nbsp;for most use cases.&nbsp;Fara1.5&nbsp;achieves&nbsp;new SOTA&nbsp;performance&nbsp;among small computer-use models and nearly doubles Fara-7B&#8217;s performance on web navigation, with&nbsp;better&nbsp;handling of forms, credentialed sites, and long-running tasks.<\/p>\n\n\n\n<p>Last November, <a href=\"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/blog\/fara-7b-an-efficient-agentic-model-for-computer-use\/\" type=\"link\" id=\"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/blog\/fara-7b-an-efficient-agentic-model-for-computer-use\/\">we released Fara-7B<\/a>, a small agentic model built for completing tasks in a web browser. It was trained using a novel synthetic data generation engine that enabled best-in-class performance. Fara1.5 is the next step in that bet: a family of three models (4B, 9B, 27B) based on Qwen 3.5, designed to close the gaps we saw in the prior release.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"what-s-new\">What&#8217;s new<\/h3>\n\n\n\n<p><strong>State-of-the-art results<\/strong>. On the popular Online-Mind2Web benchmark, which contains 300 tasks across widely used web domains, Fara1.5 sets new SOTA results for models in its size class. Fara1.5 outperforms all similarly sized models and nearly doubles the performance of Fara-7B. The larger Fara1.5-27B variant achieves more than 90% performance on the same benchmark.<\/p>\n\n\n\n<figure class=\"wp-block-image size-full\"><img loading=\"lazy\" decoding=\"async\" width=\"2560\" height=\"2048\" src=\"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-content\/uploads\/2026\/05\/Figure4-high-res_magenticlite-scaled.png\" alt=\"Figure 4 \u2013 Fara-1.5 latest results.png | A bar chart titled \"Online-Mind2Web\" measuring Task Success Rate (%, higher is better) across four models in a comparable size class. Fara (7B) scores 34%, MolmoWeb (8B) scores 35%, GUI-Owl-1.5 (8B) scores 49%, and Fara1.5 (9B) scores 63%. A curved arrow labeled \"+29 pts\" highlights the gap between GUI-Owl-1.5 and Fara1.5, illustrating that Fara1.5 achieves state-of-the-art performance among models in its weight class.\" class=\"wp-image-1171905\" srcset=\"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-content\/uploads\/2026\/05\/Figure4-high-res_magenticlite-scaled.png 2560w, https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-content\/uploads\/2026\/05\/Figure4-high-res_magenticlite-300x240.png 300w, https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-content\/uploads\/2026\/05\/Figure4-high-res_magenticlite-1024x819.png 1024w, https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-content\/uploads\/2026\/05\/Figure4-high-res_magenticlite-768x614.png 768w, https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-content\/uploads\/2026\/05\/Figure4-high-res_magenticlite-1536x1229.png 1536w, https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-content\/uploads\/2026\/05\/Figure4-high-res_magenticlite-2048x1638.png 2048w, https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-content\/uploads\/2026\/05\/Figure4-high-res_magenticlite-225x180.png 225w\" sizes=\"auto, (max-width: 2560px) 100vw, 2560px\" \/><figcaption class=\"wp-element-caption\">Figure 4.&nbsp;On the&nbsp;OnlineMind2Web&nbsp;benchmark,&nbsp;Fara\u20111.5-9B&nbsp;achieves&nbsp;state-of-the-art&nbsp;performance&nbsp;among models in its size class&nbsp;and&nbsp;substantially&nbsp;outperforms&nbsp;prior models.&nbsp;<\/figcaption><\/figure>\n\n\n\n<p><strong>Improved user experience<\/strong>. In addition to improvements on benchmarks, we improved the user experience of Fara1.5. Users should observe stronger performance on everyday tasks like filling out forms, handling logins for credentialed sites, and booking appointments. These improvements are driven by the next evolution of our FaraGen data generation pipeline. Alongside training on live websites, we also trained the model on highly realistic synthetic environments designed to simulate scenarios like logins and irreversible actions.<\/p>\n\n\n\n<p><strong>A native action space tuned for long-running tasks<\/strong>. Beyond clicks and keyboard actions, Fara1.5 has built-in tools to store key information in its context across hundreds of steps and ask the user for permission or preferences when needed, helping it stay coherent on tasks that span many minutes of real work.<\/p>\n\n\n\n<p><strong>Recalibrated critical points<\/strong>. Fara-7B was trained to detect critical points for activities like transactions, login flows, or irreversible submissions and flag them. In Fara1.5, we refined our design around critical points based on our learnings from real use, so safety triggers still occur when they should but do not block useful tasks, such as form-filling.<\/p>\n\n\n\n<figure class=\"wp-block-image size-full\"><img loading=\"lazy\" decoding=\"async\" width=\"1536\" height=\"1014\" src=\"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-content\/uploads\/2026\/05\/Figure5-Critical-points-high-res_magenticlite.png\" alt=\"Figure 5 \u2013 Critical point.png | A screenshot of Fara1.5's browser interface showing a live view of the LinkedIn sign-up and sign-in page, with fields for email and password visible. Below the browser panel, a section titled \"Input Request\" (in orange) displays the message: \"I'm at the LinkedIn login page, but I don't have the LinkedIn email and password needed to sign in. Please provide the account credentials to continue.\" This illustrates Fara1.5 pausing and prompting the user for intervention when it reaches a critical decision point requiring sensitive credentials.\" class=\"wp-image-1171909\" srcset=\"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-content\/uploads\/2026\/05\/Figure5-Critical-points-high-res_magenticlite.png 1536w, https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-content\/uploads\/2026\/05\/Figure5-Critical-points-high-res_magenticlite-300x198.png 300w, https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-content\/uploads\/2026\/05\/Figure5-Critical-points-high-res_magenticlite-1024x676.png 1024w, https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-content\/uploads\/2026\/05\/Figure5-Critical-points-high-res_magenticlite-768x507.png 768w, https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-content\/uploads\/2026\/05\/Figure5-Critical-points-high-res_magenticlite-240x158.png 240w\" sizes=\"auto, (max-width: 1536px) 100vw, 1536px\" \/><figcaption class=\"wp-element-caption\">Figure&nbsp;5.&nbsp;Fara1.5&nbsp;pauses and requests&nbsp;user intervention&nbsp;when&nbsp;it detects&nbsp;a critical point, in this case&nbsp;during a&nbsp;sign-in to&nbsp;a&nbsp;LinkedIn account using&nbsp;email&nbsp;credentials.&nbsp;<\/figcaption><\/figure>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"magenticbrain-the-orchestrator-model\">MagenticBrain: The orchestrator model<\/h3>\n\n\n\n<p>MagenticBrain&nbsp;is&nbsp;a&nbsp;14B-parameter orchestration&nbsp;model\u2014planner, coder, and delegator in one.&nbsp;Fine-tuned from Qwen 3&nbsp;14B,&nbsp;MagenticBrain&nbsp;was trained end-to-end within the&nbsp;MagenticLite&nbsp;harness&nbsp;with&nbsp;the same tool schemas&nbsp;and&nbsp;execution environment&nbsp;it&nbsp;will&nbsp;encounter&nbsp;at inference time.&nbsp;As a result,&nbsp;there&nbsp;is&nbsp;no gap between how it learned to orchestrate and how it runs.<\/p>\n\n\n\n<p>In many agentic systems, orchestration (planning and coordination) is the most reasoning-intensive component, so teams have historically relied on their most capable models for this role. Our bet is that small models can handle this role without sacrificing capability. Two design choices make that possible.<\/p>\n\n\n\n<p>The first involves combining multistep tool-calling trajectories\u2014where the model learns to pick the right tool and call it correctly\u2014with coding and terminal trajectories\u2014where the right answer is sometimes five lines of Python, not a tool call. This is paired with tight coupling between the tool format used during training and inference.<\/p>\n\n\n\n<p>The second is&nbsp;computer-use agent (CUA)&nbsp;delegation. A key part of the orchestrator&#8217;s job is knowing when&nbsp;not&nbsp;to act itself and instead handing&nbsp;off&nbsp;a&nbsp;task&nbsp;to&nbsp;Fara1.5. Our data pipeline includes explicit delegation trajectories: sequences where the orchestrator recognizes a browser or&nbsp;user interface (UI)&nbsp;task, issues a structured handoff to the CUA model, waits&nbsp;for&nbsp;the result, and resumes&nbsp;the task.&nbsp;The&nbsp;result is an orchestrator&nbsp;model&nbsp;that reasons, codes, calls tools, and delegates fluidly within a single&nbsp;14B footprint.&nbsp;We are releasing&nbsp;MagenticBrain&nbsp;which is designed&nbsp;for&nbsp;use&nbsp;with&nbsp;MagenticLite.&nbsp;<\/p>\n\n\n\n<figure class=\"wp-block-image aligncenter size-full\"><img loading=\"lazy\" decoding=\"async\" width=\"1400\" height=\"960\" src=\"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-content\/uploads\/2026\/05\/Figure-6-MagenticBrain.png\" alt=\"Figure 6 \u2013 MagenticBrain.png | A flow diagram illustrating MagenticBrain's role as an orchestration model. At the top, a box represents the user's natural-language request: \"Book me a dentist appointment Tuesday afternoon and add it to my calendar.\" An arrow points down to a central box labeled \"MagenticBrain (14B model).\" From there, four arrows branch outward to four action boxes: \"Plan steps (break it down),\" \"Call tools (calendar, web),\" \"Write code (when needed),\" and \"Delegate (hand to Fara1.5).\" A footer bar at the bottom summarizes MagenticBrain's capabilities: \"Plan \u00b7 pick tools \u00b7 code \u00b7 recover when it breaks.\"\" class=\"wp-image-1171817\" srcset=\"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-content\/uploads\/2026\/05\/Figure-6-MagenticBrain.png 1400w, https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-content\/uploads\/2026\/05\/Figure-6-MagenticBrain-300x206.png 300w, https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-content\/uploads\/2026\/05\/Figure-6-MagenticBrain-1024x702.png 1024w, https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-content\/uploads\/2026\/05\/Figure-6-MagenticBrain-768x527.png 768w, https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-content\/uploads\/2026\/05\/Figure-6-MagenticBrain-800x550.png 800w, https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-content\/uploads\/2026\/05\/Figure-6-MagenticBrain-240x165.png 240w\" sizes=\"auto, (max-width: 1400px) 100vw, 1400px\" \/><figcaption class=\"wp-element-caption\">Figure 6. MagenticBrain is a small orchestration model that can break down a natural-language request into smaller steps, select the right tools, write code when needed, and delegate browser tasks to Fara1.5.<\/figcaption><\/figure>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"the-harness-built-for-small-models\">The Harness: Built for small models<\/h3>\n\n\n\n<p>The harness combines the orchestrator and browser-use models into a single workflow. Three design choices matter most:<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>Step-by-step planning<\/strong>. The harness plans incrementally, keeping the system flexible and enabling smoother course correction and recovery throughout long-running tasks.<\/li>\n\n\n\n<li><strong>Active context management<\/strong>. Small models have smaller effective context windows and degrade faster as context grows. The harness actively curates what each model receives at each step, keeping prompts focused, surfacing only the necessary information, condensing earlier interactions into concise summaries, and offloading the rest, so the orchestrator and Fara1.5 remain effective across long tasks.<\/li>\n\n\n\n<li><strong>Delegation through subagents<\/strong>. Rather than relying on a single small model for every task, the orchestrator acts as the main agent and delegates specialized work to subagents. This means handing off browser tasks to Fara1.5. This pattern plays to the strengths of small language models by allowing each model to handle a narrower, more specialized part of the problem. It also lays the foundation for future expansion: later versions could introduce additional subagents and run them in parallel for richer, more efficient workflows.<\/li>\n<\/ul>\n\n\n\n<p>The harness preserves the human-in-the-loop guarantees from Magentic-UI 1.0. Critical points across both browser and code actions still pause for explicit user approval, and the entire system runs inside <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"https:\/\/github.com\/microsoft\/quicksand\" type=\"link\" id=\"https:\/\/github.com\/microsoft\/quicksand\" target=\"_blank\" rel=\"noopener noreferrer\">Quicksand<span class=\"sr-only\"> (opens in new tab)<\/span><\/a>, an open-source wrapper created for a QEMU-based sandbox, which isolates browser sessions and code execution from the host system.<\/p>\n\n\n\n<figure class=\"wp-block-image aligncenter size-full\"><img loading=\"lazy\" decoding=\"async\" width=\"1030\" height=\"1027\" src=\"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-content\/uploads\/2026\/05\/Figure-7-MagenticLite-Architecture-Diagram.png\" alt=\"Figure 7 \u2013 MagenticLite architecture diagram | A layered system architecture diagram for MagenticLite, organized top to bottom across four labeled sections. The topmost layer, User Interface, contains the Frontend (React SPA) with four components: Chat (conversational task input), Live Browser (noVNC stream of agent session), Approvals (human-in-the-loop gates), and Files (inputs and generated outputs). Below it, connected via WebSocket and REST, is the Orchestration layer containing the Agentic Harness (FastAPI + WebSocket). It includes four components: Orchestration (run lifecycle, streaming), Context Compaction (summarize and prune long contexts), Pause\/Resume (user-in-the-loop control), and Critical Points (detection of critical code actions), which is visually highlighted in yellow to signal its importance. The next layer is reached via a Dispatch connector and contains two parallel model components. On the left, MagenticBrain (14B model, purple) handles reasoning, coding, and delegation, with two sub-components: Reasoning Loop (think \u2192 tool \u2192 result) and Tool Dispatch (bash, edit, search, open). On the right, Fara 1.5 (9B model, teal) handles web navigation and browser use, with three sub-components: Screenshot \u2192 Action (vision-driven loop), Browser Actions (navigate, click, type, scroll), and Critical Points (forms, payments, logins). An arrow labeled \"MagenticBrain \u2192 Fara 1.5 \u00b7 delegate_cua\" connects the two models. The bottommost layer, Isolation, contains Quicksand (QEMU VM Isolation), reached via an \"Execute Inside VM\" connector. It is divided into two sections: a Sandbox with Guest Runtime (Bash, Python, Files) and CIFS Mounts (per-session workspaces); and a Browser Pool with Chromium Slots (CDP and noVNC) and Persistent Profiles (cookies and sessions).\" class=\"wp-image-1171820\" srcset=\"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-content\/uploads\/2026\/05\/Figure-7-MagenticLite-Architecture-Diagram.png 1030w, https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-content\/uploads\/2026\/05\/Figure-7-MagenticLite-Architecture-Diagram-300x300.png 300w, https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-content\/uploads\/2026\/05\/Figure-7-MagenticLite-Architecture-Diagram-1024x1021.png 1024w, https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-content\/uploads\/2026\/05\/Figure-7-MagenticLite-Architecture-Diagram-150x150.png 150w, https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-content\/uploads\/2026\/05\/Figure-7-MagenticLite-Architecture-Diagram-768x766.png 768w, https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-content\/uploads\/2026\/05\/Figure-7-MagenticLite-Architecture-Diagram-180x180.png 180w, https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-content\/uploads\/2026\/05\/Figure-7-MagenticLite-Architecture-Diagram-360x360.png 360w, https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-content\/uploads\/2026\/05\/Figure-7-MagenticLite-Architecture-Diagram-181x180.png 181w\" sizes=\"auto, (max-width: 1030px) 100vw, 1030px\" \/><figcaption class=\"wp-element-caption\">Figure 7. Overview of the MagenticLite architecture. The system uses a layered architecture spanning the front end, harness, models, and sandboxed execution environment.<\/figcaption><\/figure>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"see-it-in-action\">See it in action<\/h3>\n\n\n\n<p>MagenticLite can perform a wide range of tasks across the browser and local file system, such as filling out forms, making appointments, organizing local files, and searching for and analyzing information.<\/p>\n\n\n\n<div class=\"wp-block-columns is-layout-flex wp-container-core-columns-is-layout-9d6595d7 wp-block-columns-is-layout-flex\">\n<div class=\"wp-block-column is-layout-flow wp-block-column-is-layout-flow\">\n<figure class=\"wp-block-video\"><video height=\"960\" style=\"aspect-ratio: 960 \/ 960;\" width=\"960\" controls src=\"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-content\/uploads\/2026\/05\/MagenticLite-Fill-Expenses-Forms-Final-Demo.mp4\"><\/video><figcaption class=\"wp-element-caption\">MagenticLite | Fill expense forms demo<\/figcaption><\/figure>\n<\/div>\n\n\n\n<div class=\"wp-block-column is-layout-flow wp-block-column-is-layout-flow\">\n<figure class=\"wp-block-video\"><video height=\"960\" style=\"aspect-ratio: 960 \/ 960;\" width=\"960\" controls src=\"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-content\/uploads\/2026\/05\/MagenticLite-Finding-and-Booking-a-Restaurant-Final-Demo.mp4\"><\/video><figcaption class=\"wp-element-caption\">MagenticLite | Find and book a restaurant demo<\/figcaption><\/figure>\n<\/div>\n<\/div>\n\n\n\n<div class=\"wp-block-columns is-layout-flex wp-container-core-columns-is-layout-9d6595d7 wp-block-columns-is-layout-flex\">\n<div class=\"wp-block-column is-layout-flow wp-block-column-is-layout-flow\">\n<figure class=\"wp-block-video\"><video height=\"960\" style=\"aspect-ratio: 960 \/ 960;\" width=\"960\" controls src=\"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-content\/uploads\/2026\/05\/MagenticLite-Find-Prices-for-Recipe-Ingredients-Final-Demo.mp4\"><\/video><figcaption class=\"wp-element-caption\">MagenticLite | Find prices for recipe ingredients demo<\/figcaption><\/figure>\n<\/div>\n\n\n\n<div class=\"wp-block-column is-layout-flow wp-block-column-is-layout-flow\">\n<figure class=\"wp-block-video\"><video height=\"960\" style=\"aspect-ratio: 960 \/ 960;\" width=\"960\" controls src=\"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-content\/uploads\/2026\/05\/MagenticLite-Organize-Local-Files-Final-Demo.mp4\"><\/video><figcaption class=\"wp-element-caption\">MagenticLite | Organize local files demo<\/figcaption><\/figure>\n<\/div>\n<\/div>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"try-it-and-build-with-us\">Try it, and build with us<\/h2>\n\n\n\n<p>MagenticLite, MagenticBrain, and Fara1.5 are research releases intended to support continued exploration and development. We are releasing them to encourage experimentation, evaluation, and feedback from the broader community.<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>MagenticLite&nbsp;is&nbsp;an updated release of&nbsp;Magentic-UI,&nbsp;it\u2019s&nbsp;available on&nbsp;<a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"https:\/\/aka.ms\/magenticlite\" target=\"_blank\" rel=\"noopener noreferrer\">GitHub<span class=\"sr-only\"> (opens in new tab)<\/span><\/a>.&nbsp;<\/li>\n\n\n\n<li>MagenticBrain&nbsp;is available on <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" rel=\"noopener noreferrer\" target=\"_blank\" href=\"https:\/\/aka.ms\/MagenticBrain-foundry\">Microsoft Foundry<span class=\"sr-only\"> (opens in new tab)<\/span><\/a>.&nbsp;<\/li>\n\n\n\n<li>Fara1.5&nbsp;models are available on&nbsp;<a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" rel=\"noopener noreferrer\" target=\"_blank\" href=\"https:\/\/aka.ms\/fara-foundry\">Microsoft&nbsp;Foundry<span class=\"sr-only\"> (opens in new tab)<\/span><\/a>.&nbsp;<\/li>\n<\/ul>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"contributors\">Contributors<\/h2>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>Agentic experience<\/strong>: <a href=\"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/people\/chetan\/\">Cheng Tan<\/a>, <a href=\"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/people\/mayamurad\/\">Maya Murad<\/a>, <a href=\"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/people\/weilishi\/\">Weili Shi<\/a><\/li>\n\n\n\n<li><strong>Agentic harness<\/strong>: <a href=\"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/people\/adamfo\/\">Adam Fourney<\/a>, <a href=\"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/people\/tylerpayne\/\">Tyler Payne<\/a><\/li>\n\n\n\n<li><strong>Fara1.5<\/strong>: <a href=\"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/people\/ataymano\/\">Alexey Taymanov<\/a>, <a href=\"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/people\/andrewzhao\/\">Andrew Zhao<\/a>, <a href=\"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/people\/arrajeswaran\/\">Aravind Rajeswaran<\/a>, <a href=\"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/people\/corbyrosset\/\">Corby Rosset<\/a>, <a href=\"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/people\/hmozannar\/\">Hussein Mozannar<\/a>, <a href=\"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/people\/luizdovalle\/\">Luiz Do Valle<\/a>, <a href=\"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/people\/spwhitehead\/\">Spencer Whitehead<\/a>, <a href=\"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/people\/vivineet\/\">Vibhav Vineet<\/a>, Zach Nussbaum, <a href=\"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/people\/t-sahilgupta\/\">Sahil Gupta<\/a>, <a href=\"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/people\/luyadong\/\">Yadong Lu<\/a><\/li>\n\n\n\n<li><strong>MagenticBrain<\/strong>: <a href=\"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/people\/ahmedghoneim\/\">Ahmed Elgohary Ghoneim<\/a>, <a href=\"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/people\/akshayn\/\">Akshay Nambi<\/a>, Amir Saeidi, Caio C\u00e9sar Teodoro Mendes, <a href=\"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/people\/hbehl\/\">Harkirat Behl<\/a>, Karan Gupta, <a href=\"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/people\/pcameron\/\">Pashmina Cameron<\/a>, Pranav Vajreshwari, <a href=\"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/people\/shitals\/\">Shital Shah<\/a>, <a href=\"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/people\/yashlara\/\">Yash Lara<\/a>, <a href=\"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/people\/yashpandya\/\">Yash Pandya<\/a><\/li>\n\n\n\n<li><strong>Collaborators<\/strong>: Abhishek Gowami, <a href=\"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/people\/aswearngin\/\">Amanda Swearngin<\/a>, <a href=\"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/people\/mharrison\/\">Michael Harrison<\/a>, <a href=\"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/people\/saraabdali\/\">Sara Abdali<\/a>, Sarthak Harne, <a href=\"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/people\/vidhishab\/\">Vidhisha Balachandran<\/a><\/li>\n\n\n\n<li><strong>Project leads<\/strong>: <a href=\"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/people\/hassanam\/\">Ahmed Awadallah<\/a>, <a href=\"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/people\/raaboulh\/\">Rafah Hosn<\/a><\/li>\n\n\n\n<li><strong>Sponsors<\/strong>: <a href=\"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/people\/hassanam\/\">Ahmed Awadallah<\/a>, <a href=\"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/people\/eckamar\/\">Ece Kamar<\/a>, <a href=\"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/people\/raaboulh\/\">Rafah Hosn<\/a>, <a href=\"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/people\/samershi\/\">Saleema Amershi,<\/a> <a href=\"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/people\/shitals\/\">Shital Shah<\/a><\/li>\n<\/ul>\n\n\n\n<div style=\"height:30px\" aria-hidden=\"true\" class=\"wp-block-spacer\"><\/div>\n","protected":false},"excerpt":{"rendered":"<p>MagenticLite is an agentic system for small models that works across the browser and local file system in a single workflow. It combines specialized models and orchestration to support efficient agentic performance on everyday tasks.<\/p>\n","protected":false},"author":43868,"featured_media":1171504,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"msr-url-field":"","msr-podcast-episode":"","msrModifiedDate":"","msrModifiedDateEnabled":false,"ep_exclude_from_search":false,"_classifai_error":"","msr-author-ordering":null,"msr_hide_image_in_river":0,"footnotes":""},"categories":[1],"tags":[],"research-area":[13556],"msr-region":[],"msr-event-type":[],"msr-locale":[268875],"msr-post-option":[243984],"msr-impact-theme":[],"msr-promo-type":[],"msr-podcast-series":[],"class_list":["post-1171497","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-research-blog","msr-research-area-artificial-intelligence","msr-locale-en_us","msr-post-option-blog-homepage-featured"],"msr_event_details":{"start":"","end":"","location":""},"podcast_url":"","podcast_episode":"","msr_research_lab":[992148],"msr_impact_theme":[],"related-publications":[],"related-downloads":[],"related-videos":[],"related-academic-programs":[],"related-groups":[],"related-projects":[],"related-events":[],"related-researchers":[{"type":"guest","value":"microsoft-research-ai-frontiers","user_id":"1171853","display_name":"Microsoft Research AI Frontiers","author_link":"<a href=\"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/lab\/ai-frontiers\/people\/\" aria-label=\"Visit the profile page for Microsoft Research AI Frontiers\">Microsoft Research AI Frontiers<\/a>","is_active":true,"last_first":"AI Frontiers, Microsoft Research","people_section":0,"alias":"microsoft-research-ai-frontiers"}],"msr_type":"Post","featured_image_thumbnail":"<img width=\"960\" height=\"540\" src=\"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-content\/uploads\/2026\/05\/MagenticLite-BlogHeroFeature-1400x788-1-960x540.jpg\" class=\"img-object-cover\" alt=\"MagenticLite | Four white line icons\u2014a document in the cloud, a flowchart with three boxes, chart, and a gear\u2014displayed on a teal\u2011to\u2011green gradient background with a subtle textured pattern.\" decoding=\"async\" loading=\"lazy\" srcset=\"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-content\/uploads\/2026\/05\/MagenticLite-BlogHeroFeature-1400x788-1-960x540.jpg 960w, https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-content\/uploads\/2026\/05\/MagenticLite-BlogHeroFeature-1400x788-1-300x169.jpg 300w, https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-content\/uploads\/2026\/05\/MagenticLite-BlogHeroFeature-1400x788-1-1024x576.jpg 1024w, https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-content\/uploads\/2026\/05\/MagenticLite-BlogHeroFeature-1400x788-1-768x432.jpg 768w, https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-content\/uploads\/2026\/05\/MagenticLite-BlogHeroFeature-1400x788-1-1536x865.jpg 1536w, https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-content\/uploads\/2026\/05\/MagenticLite-BlogHeroFeature-1400x788-1-2048x1153.jpg 2048w, https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-content\/uploads\/2026\/05\/MagenticLite-BlogHeroFeature-1400x788-1-1066x600.jpg 1066w, https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-content\/uploads\/2026\/05\/MagenticLite-BlogHeroFeature-1400x788-1-655x368.jpg 655w, https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-content\/uploads\/2026\/05\/MagenticLite-BlogHeroFeature-1400x788-1-240x135.jpg 240w, https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-content\/uploads\/2026\/05\/MagenticLite-BlogHeroFeature-1400x788-1-640x360.jpg 640w, https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-content\/uploads\/2026\/05\/MagenticLite-BlogHeroFeature-1400x788-1-1280x720.jpg 1280w, https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-content\/uploads\/2026\/05\/MagenticLite-BlogHeroFeature-1400x788-1-1920x1080.jpg 1920w\" sizes=\"auto, (max-width: 960px) 100vw, 960px\" \/>","byline":"<a href=\"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/lab\/ai-frontiers\/people\/\" title=\"Go to researcher profile for Microsoft Research AI Frontiers\" aria-label=\"Go to researcher profile for Microsoft Research AI Frontiers\" data-bi-type=\"byline author\" data-bi-cN=\"Microsoft Research AI Frontiers\">Microsoft Research AI Frontiers<\/a>","formattedDate":"May 21, 2026","formattedExcerpt":"MagenticLite is an agentic system for small models that works across the browser and local file system in a single workflow. It combines specialized models and orchestration to support efficient agentic performance on everyday tasks.","locale":{"slug":"en_us","name":"English","native":"","english":"English"},"_links":{"self":[{"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/posts\/1171497","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/users\/43868"}],"replies":[{"embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/comments?post=1171497"}],"version-history":[{"count":58,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/posts\/1171497\/revisions"}],"predecessor-version":[{"id":1172993,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/posts\/1171497\/revisions\/1172993"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/media\/1171504"}],"wp:attachment":[{"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/media?parent=1171497"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/categories?post=1171497"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/tags?post=1171497"},{"taxonomy":"msr-research-area","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/research-area?post=1171497"},{"taxonomy":"msr-region","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-region?post=1171497"},{"taxonomy":"msr-event-type","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-event-type?post=1171497"},{"taxonomy":"msr-locale","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-locale?post=1171497"},{"taxonomy":"msr-post-option","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-post-option?post=1171497"},{"taxonomy":"msr-impact-theme","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-impact-theme?post=1171497"},{"taxonomy":"msr-promo-type","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-promo-type?post=1171497"},{"taxonomy":"msr-podcast-series","embeddable":true,"href":"https:\/\/cm-edgetun.pages.dev\/en-us\/research\/wp-json\/wp\/v2\/msr-podcast-series?post=1171497"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}