Microsoft puts OpenAI's GPT-6.2 Sol in Copilot worldwide as MAI models push in Microsoft said on Friday, November 20, that OpenAI's GPT-6.2 Sol is rolling out worldwide to eligible users in Copilot Chat, Copilot Cowork, Copilot Studio and Copilot in Excel, PowerPoint and Word, while the heavier GPT-6.2 Astra reaches Cowork and Studio for all eligible users but reaches Chat and the Office apps only for users in Microsoft's Frontier program. Sol costs $2 per million input tokens and $10 per million output tokens on the API versus $8 and $40 for Astra, and Microsoft's Message Center notice MC1560916 rates the admin impact as high, requiring administrators to switch on the OpenAI subprocessor setting to use Astra broadly. The rollout lands three days after Microsoft launched its own MAI Code 2 and MAI Code 2 Flash in GitHub Copilot, which Microsoft claims matches 98% of GPT-6.1 Sol at 40% of the price and delivers 96% of Claude Opus 5.5 at 5% of the cost. The newest OpenAI model is now inside Word, Excel and PowerPoint, but the stronger one is fenced off behind a paid tier. Microsoft's own MAI models launched three days earlier, so the company is selling both sides at once. Microsoft said on Friday, November 20, that OpenAI's GPT-6.2 Sol is rolling out worldwide to eligible users in Copilot Chat, Copilot Cowork, Copilot Studio, and Copilot in Excel, PowerPoint and Word, according to its Copilot blog on the Microsoft Tech Community site. GPT-6.2 Astra, the heavier model, goes to Cowork and Studio for everyone eligible, and reaches Chat and the Office apps only for users in Microsoft's Frontier program. So the line going around in digests, that every Office user got both models at once, isn't quite right. Sol is the one most people will see. Astra is the one you'll have to ask your admin about. Microsoft's own description is thin, but the pricing tells you the split. Per million tokens, Sol costs $2 for input and $10 for output on the API, while Astra costs $8 and $40, according to a Japanese-language summary of the announcement from Taiwan's Newmobilelife. Four times the price is the clue: Sol is the everyday model for drafting a memo or summarizing a thread, and Astra is meant for long, multi-step jobs where you'd pay for fewer wasted tokens. Both models run through Work IQ, which Microsoft says grounds them in a user's files, meetings, chats and business data while respecting existing permissions. Inside an Office tenant, that means the same model gives a different answer to your finance lead than to an intern: it only reads what each of them can already open. Satya Nadella says companies that rent their AI brains are making a strategic mistake they will regret https://startupfortune.com/satya-nadella-says-companies-that-rent-their-ai-brains-are-making-a-strategic-mistake-they-will-regret/ Microsoft CEO Satya Nadella has publicly warned enterprises against over-relying on foundation models from OpenAI and Anthropic, arguing that companies must build proprietary AI learning loops or risk ceding their core value to a handful of labs. The argument creates an obvious tension with Microsoft's roughly $13 billion investment in OpenAI, but... - companies building proprietary AI models https://startupfortune.com/satya-nadella-says-companies-that-rent-their-ai-brains-are-making-a-strategic-mistake-they-will-regret/ - why renting AI models fails https://startupfortune.com/satya-nadella-says-companies-that-rent-their-ai-brains-are-making-a-strategic-mistake-they-will-regret/ The admin paperwork is where the friction sits. Microsoft's Message Center notice MC1560916, summarized by MWPro, says both models use usage-based billing in supported services, and that Sol also ships with the Microsoft 365 Copilot user license. Using Astra broadly requires an administrator to switch on the OpenAI subprocessor setting. Microsoft added that OpenAI models operated by Microsoft through Azure OpenAI aren't affected by that setting. The notice rates the admin impact as high and tells IT teams to prepare support staff for usage-limit messages. Availability, Microsoft says, varies by region and organization, and the rollout was still marked in progress. If you run Copilot for a company, expect a ticket queue before expect a productivity jump. The hedge sitting next to the rollout The timing is awkward for anyone who reads Microsoft as OpenAI's loyal distributor. Three days earlier, at Ignite, Microsoft launched MAI Code 2 and MAI Code 2 Flash in GitHub Copilot. InfoWorld reported the launch as a challenge to OpenAI and Anthropic in AI coding. The Next Web carried Microsoft's own claim that MAI Code 2 matches 98% of GPT-6.1 Sol at 40% of the price, and Microsoft said it delivers 96% of Claude Opus 5.5 at 5% of the cost. Those are Microsoft's benchmarks, run on tests it picked, and the comparison prices include a launch discount that ends in May 2027. The money logic is plain. OpenAI and Microsoft formally ended their exclusive partnership and revenue-sharing arrangement on April 27, 2026, and Microsoft pays OpenAI a share of Azure AI revenue, as reported in coverage of the MAI launch. Every request served by an MAI model is revenue Microsoft keeps whole. Frankly, the two moves aren't a contradiction. Copilot in Word is a place where customers want the best-known name on the label, and Microsoft would be foolish to hold back Sol while rivals sell it. Code is different: developers compare per-task cost on a spreadsheet, and a cheaper in-house model wins there on arithmetic alone. Nadella announced GPT-6.2 for Microsoft Foundry from the same Ignite stage as the MAI Code 2 launch, which shows how deliberately the company is running both. The test for the next quarter is simple and measurable: how many Copilot requests Microsoft routes to its own models instead of OpenAI's, and whether it says so. Microsoft hasn't published that split. Also read: Lean groups certify Claude-assisted Riemann hypothesis proof from Anthropic https://startupfortune.com/lean-groups-certify-claude-assisted-riemann-hypothesis-proof-from-anthropic/ • How to Build a Golden Dataset for AI Agent Evaluation From Your Logs https://startupfortune.com/how-to-build-a-golden-dataset-for-ai-agent-evaluation-from-your-logs/ • Etsy backs off AI training on seller work after shop owners walk out https://startupfortune.com/etsy-backs-off-ai-training-on-seller-work-after-shop-owners-walk-out/ Microsoft is replacing OpenAI's image models in PowerPoint and Bing with its own MAI-Image-2 https://startupfortune.com/microsoft-is-replacing-openais-image-models-in-powerpoint-and-bing-with-its-own-mai-image-2/ Microsoft confirmed it is replacing OpenAI's image generation models in PowerPoint and Bing with its own MAI-Image-2, which ranks in the top three on the Arena.ai leaderboard. The move is part of a broader swap already underway in Excel and Outlook, as Microsoft deploys its in-house MAI model lineup to cut third-party inference costs and reduce... - Microsoft replacing OpenAI image models PowerPoint https://startupfortune.com/microsoft-is-replacing-openais-image-models-in-powerpoint-and-bing-with-its-own-mai-image-2/ - how to use MAI-Image-2 PowerPoint https://startupfortune.com/microsoft-is-replacing-openais-image-models-in-powerpoint-and-bing-with-its-own-mai-image-2/ This article is posted in AI News https://startupfortune.com/category/ai/ , check it out for more related stories. Join the discussion Open in the community → https://startupfortune.com/community/ Almost there. Sign in and your reply posts straight away.