--- title: Microsoft skipped OpenAI’s decision model and built its... description: Microsoft launched Microsoft-Decision-1 in Foundry, a decision model post-trained on Alibaba's Qwen3.5-9B rather than an OpenAI base, three days after OpenAI... canonical: https://daily.dev/posts/microsoft-skipped-openai-s-decision-model-and-built-its-own-on-alibaba-s-qwen-wpidg8er9 twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: Microsoft skipped OpenAI’s decision model and built its own on Alibaba’s Qwen | daily.dev og:description: Microsoft launched Microsoft-Decision-1 in Foundry, a decision model post-trained on Alibaba's Qwen3.5-9B rather than an OpenAI base, three days after OpenAI... og:url: https://daily.dev/posts/microsoft-skipped-openai-s-decision-model-and-built-its-own-on-alibaba-s-qwen-wpidg8er9 og:image: https://api.daily.dev/og/posts/wPiDg8eR9.png og:image:alt: Microsoft skipped OpenAI’s decision model and built its own on Alibaba’s Qwen og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # Microsoft skipped OpenAI’s decision model and built its own on Alibaba’s Qwen **[The New Stack](https://daily.dev/sources/newstack)** · 5 min read · 0 upvotes · 0 comments ## Summary Microsoft launched Microsoft-Decision-1 in Foundry, a decision model post-trained on Alibaba's Qwen3.5-9B rather than an OpenAI base, three days after OpenAI opened its own Decisions API to all developers. Priced at $0.042 per million input tokens with free output, matching TypeSafe's Jev pricing, it arrived the same day TypeSafe announced an $870 million Series A at a $7.5 billion valuation. Microsoft is already using it internally across Xbox Research, Copilot, on-call incident response, and Microsoft Discovery, claiming it's over 14x faster than GPT-6 Sol for some tasks. The model lacks image support (unlike OpenAI's and Cloudflare's offerings), has no announced open weights, and its compatibility with the emerging System One API standard is unconfirmed. Adversarial testing showed Decision-1's answers changed only 1.3% under perturbation, though it hasn't been tested against the more aggressive JevOut attack methodology that flipped the majority of a rival model's correct decisions. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### What base model did Microsoft use to build Decision-1 instead of an OpenAI model? Microsoft post-trained Decision-1 on Alibaba's Qwen3.5-9B rather than on any OpenAI model, despite OpenAI being its partner. Microsoft plans to rebase the model later on its own MAI models as well as OpenAI's. Decision-1 launched in Foundry priced at $0.042 per million input tokens with free output, matching TypeSafe's Jev pricing. _Developers choosing a decision-model provider can track how pricing and base models compare on daily.dev._ ### Does Microsoft's Decision-1 model support image inputs like OpenAI's Decisions API? No, Decision-1 only accepts up to 32,768 tokens of text and returns JSON; it does not support images. This puts it behind OpenAI's Decisions API and Cloudflare's Clef, which uses a vision encoder for multimodal decision-making. Microsoft also has not announced open weights for Decision-1, unlike Cloudflare, which released Clef under Apache 2.0. _Teams weighing multimodal support when picking a decision model can follow these tradeoffs on daily.dev._ ### How robust is Microsoft's Decision-1 model against adversarial perturbations compared to other decision models? Microsoft tested Decision-1 against eight kinds of perturbations, including reordered options and paraphrased descriptions, which changed only 1.3% of its answers on average. By contrast, the JevOut preprint found that similar adversarial additions flipped 312 of 508 initially correct decisions from TypeSafe's Jev model, with three other scoring systems showing flip rates between 64.9% and 73.2%. Decision-1 itself was not tested using the JevOut methodology. _Developers building agents that rely on calibrated confidence scores can follow adversarial robustness results on daily.dev._ ## Similar posts on daily.dev - [Microsoft just admitted its biggest AI mistake — and spent $2.5 billion fixing it](https://daily.dev/posts/microsoft-just-admitted-its-biggest-ai-mistake-and-spent-2-5-billion-fixing-it-cxuqalrpi) · The New Stack · 0 upvotes · 0 comments --- Tags: [#ai-agents](https://daily.dev/tags/ai-agents), [#openai](https://daily.dev/tags/openai), [#qwen](https://daily.dev/tags/qwen) [View this post on daily.dev](https://daily.dev/posts/microsoft-skipped-openai-s-decision-model-and-built-its-own-on-alibaba-s-qwen-wpidg8er9) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"Microsoft skipped OpenAI’s decision model and built its own on Alibaba’s Qwen","url":"https://daily.dev/posts/microsoft-skipped-openai-s-decision-model-and-built-its-own-on-alibaba-s-qwen-wpidg8er9","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/microsoft-skipped-openai-s-decision-model-and-built-its-own-on-alibaba-s-qwen-wpidg8er9"},"datePublished":"2026-10-10T13:08:28.806Z","dateModified":"2026-10-11T12:38:57.534Z","description":"Microsoft launched Microsoft-Decision-1 in Foundry, a decision model post-trained on Alibaba's Qwen3.5-9B rather than an OpenAI base, three days after OpenAI...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/22003d1bd37a18a4528eca98492548f5?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/22003d1bd37a18a4528eca98492548f5?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"The New Stack","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"The New Stack","logo":"https://media.daily.dev/image/upload/t_logo,f_auto/v1/logos/newstack","url":"https://daily.dev/sources/newstack"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/microsoft-skipped-openai-s-decision-model-and-built-its-own-on-alibaba-s-qwen-wpidg8er9","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":0},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"ai-agents,openai,qwen","timeRequired":"PT5M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"The New Stack","item":"https://daily.dev/sources/newstack"},{"@type":"ListItem","position":3,"name":"Microsoft skipped OpenAI’s decision model and built its own on Alibaba’s Qwen"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/microsoft-skipped-openai-s-decision-model-and-built-its-own-on-alibaba-s-qwen-wpidg8er9#faq","mainEntity":[{"@type":"Question","name":"What base model did Microsoft use to build Decision-1 instead of an OpenAI model?","acceptedAnswer":{"@type":"Answer","text":"Microsoft post-trained Decision-1 on Alibaba's Qwen3.5-9B rather than on any OpenAI model, despite OpenAI being its partner. Microsoft plans to rebase the model later on its own MAI models as well as OpenAI's. Decision-1 launched in Foundry priced at $0.042 per million input tokens with free output, matching TypeSafe's Jev pricing. Developers choosing a decision-model provider can track how pricing and base models compare on daily.dev."}},{"@type":"Question","name":"Does Microsoft's Decision-1 model support image inputs like OpenAI's Decisions API?","acceptedAnswer":{"@type":"Answer","text":"No, Decision-1 only accepts up to 32,768 tokens of text and returns JSON; it does not support images. This puts it behind OpenAI's Decisions API and Cloudflare's Clef, which uses a vision encoder for multimodal decision-making. Microsoft also has not announced open weights for Decision-1, unlike Cloudflare, which released Clef under Apache 2.0. Teams weighing multimodal support when picking a decision model can follow these tradeoffs on daily.dev."}},{"@type":"Question","name":"How robust is Microsoft's Decision-1 model against adversarial perturbations compared to other decision models?","acceptedAnswer":{"@type":"Answer","text":"Microsoft tested Decision-1 against eight kinds of perturbations, including reordered options and paraphrased descriptions, which changed only 1.3% of its answers on average. By contrast, the JevOut preprint found that similar adversarial additions flipped 312 of 508 initially correct decisions from TypeSafe's Jev model, with three other scoring systems showing flip rates between 64.9% and 73.2%. Decision-1 itself was not tested using the JevOut methodology. Developers building agents that rely on calibrated confidence scores can follow adversarial robustness results on daily.dev."}}]} ```