{"id":1159707,"date":"2026-01-15T06:00:00","date_gmt":"2026-01-15T14:00:00","guid":{"rendered":"https:\/\/www.microsoft.com\/en-us\/research\/?p=1159707"},"modified":"2026-01-15T06:00:04","modified_gmt":"2026-01-15T14:00:04","slug":"optimind-a-small-language-model-with-optimization-expertise","status":"publish","type":"post","link":"https:\/\/www.microsoft.com\/en-us\/research\/blog\/optimind-a-small-language-model-with-optimization-expertise\/","title":{"rendered":"OptiMind: A small language model with optimization expertise"},"content":{"rendered":"\n<figure class=\"wp-block-image size-full\"><img loading=\"lazy\" decoding=\"async\" width=\"1400\" height=\"788\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2026\/01\/OptiMind-BlogHeroFeature-1400x788-2.jpg\" alt=\"A flowchart with three horizontal sections on a blue-to-green gradient background. The first section, labeled \u201cClassification,\u201d shows icons of a computer, an arrow pointing to a robot face, and another arrow pointing to a box labeled \u201cTSP.\u201d The second section, labeled \u201cInference,\u201d displays a robot icon connected by arrows to two document icons, one of which includes a magnifying glass. The third section, labeled \u201cTest-time scaling,\u201d shows a document with a checkmark connected by an arrow to a circular refresh icon. Arrows indicate the flow between sections, starting from Classification to Inference and then to Test-time scaling.\" class=\"wp-image-1159915\" srcset=\"https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2026\/01\/OptiMind-BlogHeroFeature-1400x788-2.jpg 1400w, https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2026\/01\/OptiMind-BlogHeroFeature-1400x788-2-300x169.jpg 300w, https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2026\/01\/OptiMind-BlogHeroFeature-1400x788-2-1024x576.jpg 1024w, https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2026\/01\/OptiMind-BlogHeroFeature-1400x788-2-768x432.jpg 768w, https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2026\/01\/OptiMind-BlogHeroFeature-1400x788-2-1066x600.jpg 1066w, https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2026\/01\/OptiMind-BlogHeroFeature-1400x788-2-655x368.jpg 655w, https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2026\/01\/OptiMind-BlogHeroFeature-1400x788-2-240x135.jpg 240w, https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2026\/01\/OptiMind-BlogHeroFeature-1400x788-2-640x360.jpg 640w, https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2026\/01\/OptiMind-BlogHeroFeature-1400x788-2-960x540.jpg 960w, https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2026\/01\/OptiMind-BlogHeroFeature-1400x788-2-1280x720.jpg 1280w\" sizes=\"auto, (max-width: 1400px) 100vw, 1400px\" \/><\/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 is-layout-flex wp-container-core-columns-is-layout-9d6595d7 wp-block-columns-is-layout-flex\">\n<div class=\"wp-block-column is-style-default mb-10 pb-1 pr-1 is-layout-flow wp-block-column-is-layout-flow\" style=\"box-shadow:var(--wp--preset--shadow--outlined)\">\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>Many real-world business problems can benefit from optimization, but translating decisions, constraints, and goals from natural language into optimization algorithms is slow.<\/li>\n\n\n\n<li>OptiMind is a small language model designed to convert business problems described in natural language into the mathematical formulations needed by optimization software.<\/li>\n\n\n\n<li>OptiMind&nbsp;is trained on&nbsp;a carefully curated, expert-aligned dataset&nbsp;and applies domain-specific hints and self-checks at inference time, improving&nbsp;its&nbsp;accuracy.<\/li>\n\n\n\n<li>OptiMind&nbsp;matches or exceeds the performance of much larger systems,&nbsp;can&nbsp;run&nbsp;locally to protect sensitive data,&nbsp;produces&nbsp;more reliable formulations, and&nbsp;reduces the time and&nbsp;expertise&nbsp;needed to prepare optimization models.<\/li>\n<\/ul>\n<\/div>\n<\/div>\t\t<\/div>\n\t<\/div>\n\n\t<\/div>\n\n\n\n<p>Enterprises across industries, from energy to finance, use optimization models to plan complex operations like supply chains and logistics. These models work by defining three elements: the choices that can be made (such as production quantities or delivery routes), the rules and limits those choices must follow, and the goal, whether that\u2019s minimizing costs, meeting customer demand, or improving efficiency.<\/p>\n\n\n\n<p>Over the past few decades,&nbsp;many&nbsp;businesses have shifted&nbsp;from&nbsp;judgment-based decision-making to data-driven&nbsp;approaches,&nbsp;leading to&nbsp;major&nbsp;efficiency gains and cost&nbsp;savings. Advances in AI promise to accelerate this shift even further, potentially cutting decision times from days to minutes while delivering better results.<\/p>\n\n\n\n<p>In practice, however, turning real-world business problems into a form that optimization software can understand is challenging. This translation process requires expressing decisions, constraints, and objectives in mathematical terms. The work demands specialized&nbsp;expertise,&nbsp;and&nbsp;it&nbsp;can take anywhere from&nbsp;one&nbsp;day to several weeks&nbsp;to solve&nbsp;complex problems.&nbsp;<\/p>\n\n\n\n<p>To address this challenge, we\u2019re introducing <a href=\"https:\/\/www.microsoft.com\/en-us\/research\/publication\/optimind-teaching-llms-to-think-like-optimization-experts\/\">OptiMind<\/a>, a small language model designed to convert problems described in plain language into the mathematical formulations that optimization software needs. Built on a 20-billion parameter model, OptiMind is compact by today\u2019s standards yet matches the performance of larger, more complex systems. Its modest size means it can run locally on users\u2019 devices, enabling fast iteration while keeping sensitive business data on users\u2019 devices rather than transmitting it to external servers.<\/p>\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=\"999693\">\n\t\t\n\n\t\t<p class=\"msr-promo__label text-gray-800 text-center text-uppercase\">\n\t\t<span class=\"px-4 bg-white display-inline-block font-weight-semibold small\">Spotlight: Event Series<\/span>\n\t<\/p>\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:\/\/www.microsoft.com\/en-us\/research\/event\/microsoft-research-forum\/?OCID=msr_researchforum_MCR_Blog_Promo\" aria-label=\"Microsoft Research Forum\" data-bi-cN=\"Microsoft Research Forum\" target=\"_blank\">\n\t\t\t\t\t<img decoding=\"async\" class=\"w-100 display-block\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2025\/05\/Research-Forum-hero_1400x788.jpg\" alt=\"Research Forum | abstract background with colorful hexagons\" \/>\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\">Microsoft Research Forum<\/h2>\n\t\t\t\t\n\t\t\t\t\t\t\t\t<p id=\"microsoft-research-forum\" class=\"large\">Join us for a continuous exchange of ideas about research in the era of general AI. Watch the first four episodes on demand.<\/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:\/\/www.microsoft.com\/en-us\/research\/event\/microsoft-research-forum\/?OCID=msr_researchforum_MCR_Blog_Promo\" aria-describedby=\"microsoft-research-forum\" class=\"btn btn-brand glyph-append glyph-append-chevron-right\" data-bi-cN=\"Microsoft Research Forum\" target=\"_blank\">\n\t\t\t\t\t\t\tWatch on-demand\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=\"how-it-works\">How it works<strong><\/strong><\/h2>\n\n\n\n<p>OptiMind incorporates knowledge from optimization experts both during training and when it\u2019s being used to improve formulation accuracy at scale. Three stages enable this: domain-specific hints improve training data quality, the model undergoes fine-tuning, and expert reasoning guides the model as it works.<\/p>\n\n\n\n<figure class=\"wp-block-image aligncenter size-full\"><img loading=\"lazy\" decoding=\"async\" width=\"2560\" height=\"750\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2026\/01\/OptiMind-1-scaled.jpg\" alt=\"The image illustrates a linear programming model for a manufacturing facility, detailing the production quantities, setup indicators, and inventory levels for different products over a six-month period, aiming to optimize costs.\" class=\"wp-image-1160007\" srcset=\"https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2026\/01\/OptiMind-1-scaled.jpg 2560w, https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2026\/01\/OptiMind-1-300x88.jpg 300w, https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2026\/01\/OptiMind-1-1024x300.jpg 1024w, https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2026\/01\/OptiMind-1-768x225.jpg 768w, https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2026\/01\/OptiMind-1-1536x450.jpg 1536w, https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2026\/01\/OptiMind-1-2048x600.jpg 2048w, https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2026\/01\/OptiMind-1-240x70.jpg 240w\" sizes=\"auto, (max-width: 2560px) 100vw, 2560px\" \/><figcaption class=\"wp-element-caption\">Figure 1.&nbsp;From problem description to solution&nbsp;<\/figcaption><\/figure>\n\n\n\n<p>One of the central challenges&nbsp;in&nbsp;developing&nbsp;OptiMind was the poor quality of existing public datasets for optimization problems. Many examples were incomplete or&nbsp;contained incorrect solutions. To address this, we developed a systematic approach that combines automation with expert review. It organizes problems into well-known categories, such as scheduling or routing, and identifies common error patterns within each. Using these insights, we generated expert-verified &#8220;hints&#8221; to guide the process, enabling the system to regenerate higher-quality solutions and filter out unsolvable examples (Figure 2). The result is a training dataset that more accurately reflects how optimization experts structure problems.<\/p>\n\n\n\n<figure class=\"wp-block-image aligncenter size-full is-resized\"><img loading=\"lazy\" decoding=\"async\" width=\"1648\" height=\"638\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2026\/01\/training-data-cleaning.png\" alt=\"Process for correcting training data\" class=\"wp-image-1159877\" style=\"aspect-ratio:2.58319131760696;width:526px;height:auto\" srcset=\"https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2026\/01\/training-data-cleaning.png 1648w, https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2026\/01\/training-data-cleaning-300x116.png 300w, https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2026\/01\/training-data-cleaning-1024x396.png 1024w, https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2026\/01\/training-data-cleaning-768x297.png 768w, https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2026\/01\/training-data-cleaning-1536x595.png 1536w, https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2026\/01\/training-data-cleaning-240x93.png 240w\" sizes=\"auto, (max-width: 1648px) 100vw, 1648px\" \/><figcaption class=\"wp-element-caption\">Figure 2. Process for correcting training data<\/figcaption><\/figure>\n\n\n\n<p>Using this refined dataset, we applied supervised fine-tuning to the base model. Rather than simply generating code, we trained OptiMind to produce structured mathematical formulations alongside intermediate reasoning steps, helping it avoid the common mistakes found in earlier datasets.<\/p>\n\n\n\n<p>When in use, the model&#8217;s reliability further improves. When given a new problem, OptiMind first classifies it into a category, such as scheduling or network design. It then applies expert hints relevant to that type of problem, which act as reminders to check for errors before generating a solution. For particularly challenging problems, the system generates multiple solutions and either selects the most&nbsp;frequently&nbsp;occurring one or uses feedback to refine its response. This approach increases accuracy without requiring a larger model, as illustrated in Figure 3.<\/p>\n\n\n\n<figure class=\"wp-block-image aligncenter size-full is-resized\"><img loading=\"lazy\" decoding=\"async\" width=\"2057\" height=\"817\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2026\/01\/inference-pipeline.png\" alt=\"OptiMind\u2019s inference process\" class=\"wp-image-1159878\" style=\"aspect-ratio:2.5178126502600864;width:770px;height:auto\" srcset=\"https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2026\/01\/inference-pipeline.png 2057w, https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2026\/01\/inference-pipeline-300x119.png 300w, https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2026\/01\/inference-pipeline-1024x407.png 1024w, https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2026\/01\/inference-pipeline-768x305.png 768w, https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2026\/01\/inference-pipeline-1536x610.png 1536w, https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2026\/01\/inference-pipeline-2048x813.png 2048w, https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2026\/01\/inference-pipeline-240x95.png 240w\" sizes=\"auto, (max-width: 2057px) 100vw, 2057px\" \/><figcaption class=\"wp-element-caption\">Figure 3. OptiMind\u2019s inference process<\/figcaption><\/figure>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"evaluation\">Evaluation<\/h2>\n\n\n\n<p>To test the system, we turned to three widely used public benchmarks that&nbsp;represent&nbsp;some of the most complex formulation tasks in the field. On closer inspection, we discovered that&nbsp;30 to 50 percent&nbsp;of the original test data was flawed.&nbsp;After manually correcting the issues, OptiMind improved accuracy by approximately 10 percent over the base model. Figure 4 and Table 1 show detailed comparisons: OptiMind outperformed other open-source models under 32 billion parameters and, when combined with expert hints and correction strategies, matched or exceeded the performance of current leading models.<\/p>\n\n\n\n<figure class=\"wp-block-image aligncenter size-full is-resized\"><img loading=\"lazy\" decoding=\"async\" width=\"1231\" height=\"457\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2026\/01\/main_barplot_new-1.png\" alt=\"Average accuracy percentages over all models.\" class=\"wp-image-1160042\" style=\"aspect-ratio:2.461748867424566;width:850px;height:auto\" srcset=\"https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2026\/01\/main_barplot_new-1.png 1231w, https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2026\/01\/main_barplot_new-1-300x111.png 300w, https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2026\/01\/main_barplot_new-1-1024x380.png 1024w, https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2026\/01\/main_barplot_new-1-768x285.png 768w, https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2026\/01\/main_barplot_new-1-240x89.png 240w\" sizes=\"auto, (max-width: 1231px) 100vw, 1231px\" \/><figcaption class=\"wp-element-caption\">Figure 4. Average accuracy percentages over all models.<\/figcaption><\/figure>\n\n\n\n<figure class=\"wp-block-image aligncenter size-full\"><img loading=\"lazy\" decoding=\"async\" width=\"2122\" height=\"922\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2026\/01\/main-results-table.png\" alt=\"Performance of all models on corrected benchmark datasets\" class=\"wp-image-1159880\" srcset=\"https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2026\/01\/main-results-table.png 2122w, https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2026\/01\/main-results-table-300x130.png 300w, https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2026\/01\/main-results-table-1024x445.png 1024w, https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2026\/01\/main-results-table-768x334.png 768w, https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2026\/01\/main-results-table-1536x667.png 1536w, https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2026\/01\/main-results-table-2048x890.png 2048w, https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2026\/01\/main-results-table-240x104.png 240w\" sizes=\"auto, (max-width: 2122px) 100vw, 2122px\" \/><figcaption class=\"wp-element-caption\">Table 1. Performance of all models on corrected benchmark datasets<\/figcaption><\/figure>\n\n\n\n<p>OptiMind is more reliable than other models because it learns from higher-quality, domain-aligned data. And by correcting errors and inconsistencies in standard datasets, we significantly reduced the model&#8217;s tendency to hallucinate relative&nbsp;to the base and comparison models.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"looking-forward\">Looking forward<strong><\/strong><\/h2>\n\n\n\n<p>While supervised fine-tuning has provided a strong foundation, we are exploring reinforcement learning to further refine OptiMind&#8217;s reasoning capabilities. We\u2019re also investigating automated frameworks that would allow LLMs to generate their own expert hints, enabling continuous autonomous improvement. Additionally, we are working with Microsoft product teams and industry collaborators to expand OptiMind\u2019s utility, adding support for more programming languages and a variety of input formats, including Excel and other widely used tools.<\/p>\n\n\n\n<p>We&#8217;re releasing OptiMind as an experimental model to gather community feedback and inform future development. The model is available through <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"https:\/\/aka.ms\/OptiMindCatalog\" target=\"_blank\" rel=\"noopener noreferrer\">Microsoft Foundry<span class=\"sr-only\"> (opens in new tab)<\/span><\/a> and <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"https:\/\/aka.ms\/OptiMindHF\" target=\"_blank\" rel=\"noopener noreferrer\">Hugging Face<span class=\"sr-only\"> (opens in new tab)<\/span><\/a>, and we\u2019ve open-sourced the benchmarks and data-processing procedures on <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"https:\/\/aka.ms\/OptiGuideGithub\" target=\"_blank\" rel=\"noopener noreferrer\">GitHub<span class=\"sr-only\"> (opens in new tab)<\/span><\/a> to support more reliable evaluation across the field. We welcome feedback through <a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"https:\/\/aka.ms\/OptiGuideGithub\" target=\"_blank\" rel=\"noopener noreferrer\">GitHub<span class=\"sr-only\"> (opens in new tab)<\/span><\/a>, and invite those interested in shaping the future of optimization to\u00a0apply for one of our\u00a0<a href=\"https:\/\/www.microsoft.com\/en-us\/research\/careers\/\">open roles<\/a>.<\/p>\n\n\n\n<p><\/p>\n","protected":false},"excerpt":{"rendered":"<p>OptiMind is a small language model that converts business operation challenges, described naturally, into mathematical formulations that optimization software can solve. It reduces formulation time & errors & enables fast, privacy-preserving local use.<\/p>\n","protected":false},"author":43518,"featured_media":1159912,"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":null,"footnotes":""},"categories":[1],"tags":[],"research-area":[13556,13546],"msr-region":[],"msr-event-type":[],"msr-locale":[268875],"msr-post-option":[269148,243984,269142],"msr-impact-theme":[],"msr-promo-type":[],"msr-podcast-series":[],"class_list":["post-1159707","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-research-blog","msr-research-area-artificial-intelligence","msr-research-area-computational-sciences-mathematics","msr-locale-en_us","msr-post-option-approved-for-river","msr-post-option-blog-homepage-featured","msr-post-option-include-in-river"],"msr_event_details":{"start":"","end":"","location":""},"podcast_url":"","podcast_episode":"","msr_research_lab":[199565],"msr_impact_theme":[],"related-publications":[],"related-downloads":[],"related-videos":[],"related-academic-programs":[],"related-groups":[569136],"related-projects":[1112937],"related-events":[],"related-researchers":[{"type":"guest","value":"xinzhi-zhang","user_id":"1159886","display_name":"Xinzhi Zhang","author_link":"<a href=\"https:\/\/www.linkedin.com\/in\/xinzhi-zhang-55a8662a2\/\" aria-label=\"Visit the profile page for Xinzhi Zhang\">Xinzhi Zhang<\/a>","is_active":true,"last_first":"Zhang, Xinzhi","people_section":0,"alias":"xinzhi-zhang"},{"type":"guest","value":"zeyi-chen","user_id":"1143453","display_name":"Zeyi Chen","author_link":"<a href=\"https:\/\/www.linkedin.com\/in\/zeyi-chen-a429a2218\/\" aria-label=\"Visit the profile page for Zeyi Chen\">Zeyi Chen<\/a>","is_active":true,"last_first":"Chen, Zeyi","people_section":0,"alias":"zeyi-chen"},{"type":"guest","value":"humishka-hope","user_id":"1160031","display_name":"Humishka Hope","author_link":"<a href=\"https:\/\/www.linkedin.com\/in\/humishka-zope\/\" aria-label=\"Visit the profile page for Humishka Hope\">Humishka Hope<\/a>","is_active":true,"last_first":"Hope, Humishka","people_section":0,"alias":"humishka-hope"},{"type":"user_nicename","value":"Hugo Barbalho","user_id":40744,"display_name":"Hugo Barbalho","author_link":"<a href=\"https:\/\/www.microsoft.com\/en-us\/research\/people\/hugobarbalho\/\" aria-label=\"Visit the profile page for Hugo Barbalho\">Hugo Barbalho<\/a>","is_active":false,"last_first":"Barbalho, Hugo","people_section":0,"alias":"hugobarbalho"},{"type":"user_nicename","value":"Konstantina Mellou","user_id":38874,"display_name":"Konstantina Mellou","author_link":"<a href=\"https:\/\/www.microsoft.com\/en-us\/research\/people\/kmellou\/\" aria-label=\"Visit the profile page for Konstantina Mellou\">Konstantina Mellou<\/a>","is_active":false,"last_first":"Mellou, Konstantina","people_section":0,"alias":"kmellou"},{"type":"user_nicename","value":"Marco Molinaro","user_id":42204,"display_name":"Marco Molinaro","author_link":"<a href=\"https:\/\/www.microsoft.com\/en-us\/research\/people\/mmolinaro\/\" aria-label=\"Visit the profile page for Marco Molinaro\">Marco Molinaro<\/a>","is_active":false,"last_first":"Molinaro, Marco","people_section":0,"alias":"mmolinaro"},{"type":"user_nicename","value":"Janardhan (Jana) Kulkarni","user_id":32147,"display_name":"Janardhan (Jana) Kulkarni","author_link":"<a href=\"https:\/\/www.microsoft.com\/en-us\/research\/people\/jakul\/\" aria-label=\"Visit the profile page for Janardhan (Jana) Kulkarni\">Janardhan (Jana) Kulkarni<\/a>","is_active":false,"last_first":"Kulkarni, Janardhan (Jana)","people_section":0,"alias":"jakul"},{"type":"user_nicename","value":"Ishai Menache","user_id":32116,"display_name":"Ishai Menache","author_link":"<a href=\"https:\/\/www.microsoft.com\/en-us\/research\/people\/ishai\/\" aria-label=\"Visit the profile page for Ishai Menache\">Ishai Menache<\/a>","is_active":false,"last_first":"Menache, Ishai","people_section":0,"alias":"ishai"},{"type":"user_nicename","value":"Sirui Li","user_id":43857,"display_name":"Sirui Li","author_link":"<a href=\"https:\/\/www.microsoft.com\/en-us\/research\/people\/siruili\/\" aria-label=\"Visit the profile page for Sirui Li\">Sirui Li<\/a>","is_active":false,"last_first":"Li, Sirui","people_section":0,"alias":"siruili"}],"msr_type":"Post","featured_image_thumbnail":"<img width=\"960\" height=\"540\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2026\/01\/OptiMind-BlogHeroFeature-1400x788-1-960x540.jpg\" class=\"img-object-cover\" alt=\"A flowchart with three horizontal sections on a blue-to-green gradient background. The first section, labeled \u201cClassification,\u201d shows icons of a computer, an arrow pointing to a robot face, and another arrow pointing to a box labeled \u201cTSP.\u201d The second section, labeled \u201cInference,\u201d displays a robot icon connected by arrows to two document icons, one of which includes a magnifying glass. The third section, labeled \u201cTest-time scaling,\u201d shows a document with a checkmark connected by an arrow to a circular refresh icon. Arrows indicate the flow between sections, starting from Classification to Inference and then to Test-time scaling.\" decoding=\"async\" loading=\"lazy\" srcset=\"https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2026\/01\/OptiMind-BlogHeroFeature-1400x788-1-960x540.jpg 960w, https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2026\/01\/OptiMind-BlogHeroFeature-1400x788-1-300x169.jpg 300w, https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2026\/01\/OptiMind-BlogHeroFeature-1400x788-1-1024x576.jpg 1024w, https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2026\/01\/OptiMind-BlogHeroFeature-1400x788-1-768x432.jpg 768w, https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2026\/01\/OptiMind-BlogHeroFeature-1400x788-1-1066x600.jpg 1066w, https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2026\/01\/OptiMind-BlogHeroFeature-1400x788-1-655x368.jpg 655w, https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2026\/01\/OptiMind-BlogHeroFeature-1400x788-1-240x135.jpg 240w, https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2026\/01\/OptiMind-BlogHeroFeature-1400x788-1-640x360.jpg 640w, https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2026\/01\/OptiMind-BlogHeroFeature-1400x788-1-1280x720.jpg 1280w, https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2026\/01\/OptiMind-BlogHeroFeature-1400x788-1.jpg 1400w\" sizes=\"auto, (max-width: 960px) 100vw, 960px\" \/>","byline":"","formattedDate":"January 15, 2026","formattedExcerpt":"OptiMind is a small language model that converts business operation challenges, described naturally, into mathematical formulations that optimization software can solve. It reduces formulation time &amp; errors &amp; enables fast, privacy-preserving local use.","locale":{"slug":"en_us","name":"English","native":"","english":"English"},"_links":{"self":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/posts\/1159707","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/users\/43518"}],"replies":[{"embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/comments?post=1159707"}],"version-history":[{"count":59,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/posts\/1159707\/revisions"}],"predecessor-version":[{"id":1160170,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/posts\/1159707\/revisions\/1160170"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/media\/1159912"}],"wp:attachment":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/media?parent=1159707"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/categories?post=1159707"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/tags?post=1159707"},{"taxonomy":"msr-research-area","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/research-area?post=1159707"},{"taxonomy":"msr-region","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-region?post=1159707"},{"taxonomy":"msr-event-type","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-event-type?post=1159707"},{"taxonomy":"msr-locale","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-locale?post=1159707"},{"taxonomy":"msr-post-option","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-post-option?post=1159707"},{"taxonomy":"msr-impact-theme","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-impact-theme?post=1159707"},{"taxonomy":"msr-promo-type","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-promo-type?post=1159707"},{"taxonomy":"msr-podcast-series","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-podcast-series?post=1159707"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}