![\"After](\"https:\/\/static.toiimg.com\/thumb\/msid-132980408,imgsize-789835,width-400,height-225,resizemode-4\/microsofts-new-internal-memo-makes-openais-gpt-56-the-default-model-for-its-engineers-replacing-a-setup-that-had-been-quietly-routing-github-copilot-work-to-anthropics-claude.jpg\" "\\"Microsoft's")
<\/div>\n<\/div>\n<\/div>\nMicrosoft spent May telling its Windows, Office and Surface engineers to wind down Claude Code and move to GitHub Copilot CLI before the fiscal year closed on June 30. That looked like a licensing decision, and an expensive one for Anthropic. This week’s internal memo suggests the harder blow was still coming.Jay Parikh, the executive vice president who runs Microsoft’s CoreAI organisation, wrote to his teams that “tokenmaxxing is not what we are optimizing for,” asked engineers to review their own AI spending, and made OpenAI’s GPT-5.6 the default model for internal use. 404 Media first reported the memo, which lands at a company that has just posted another record quarter and is still worried enough about its own AI bill to put a ceiling on it.<\/p>\n The default-model line is the part Anthropic will feel. According to reporting by Alex Heath, Microsoft’s internal GitHub Copilot configuration previously ran on an auto-router that leaned mostly on Anthropic’s models. In practice, thousands of Microsoft engineers were writing code with Claude tokens without ever choosing Claude, and Microsoft was paying for it. Cancelling Claude Code took away a product people had asked for. Changing the router takes away the traffic nobody had to ask for.Parikh framed it as efficiency rather than divorce. Shifting more workloads to OpenAI models, he wrote, helps the company get greater value from its token investment. He also stressed that Microsoft is not chasing lower usage: “We are not optimizing for fewer tokens. We are optimizing for more impact per token.”<\/p>\n Since July, every Microsoft division has carried an AI token budget target, and employees can look up their individual spend on an internal dashboard. The guidelines seen by 404 Media note that many engineers were burning through anywhere from a few hundred to a few thousand dollars in tokens each month.Microsoft is late to this. Uber’s engineering organisation exhausted its entire planned 2026 AI coding budget in four months. Satya Nadella<\/a> himself admitted at a live Hard Fork taping in June that plenty of tokenmaxxing was happening inside Microsoft, called the habit addictive, then delivered the line that now reads like policy: don’t use frontier models for non-frontier problems.<\/p>\n None of this ends the relationship. Claude models remain reachable through Copilot CLI, Microsoft Foundry customers still get Sonnet, Opus and Haiku under the November deal, and Microsoft has continued to favour Anthropic inside Microsoft 365 Copilot for tasks where it outperforms OpenAI. Microsoft even worked with Anthropic to bring Claude Cowork technology into Microsoft 365 Copilot.But defaults are where volume lives. Anthropic’s business has been built substantially on enterprise coding workloads, and its largest customer just rewired its internal plumbing to send those workloads elsewhere by default. Rajesh Jha’s May memo said Copilot CLI gave Microsoft a product it could shape directly with GitHub. Parikh’s memo explains the other half of that sentence. Microsoft wanted a tool it controlled, and a bill it could predict, and Claude was expensive on both counts.<\/div>\n Microsoft’s new internal memo makes OpenAI’s GPT-5.6 the default model for its engineers, replacing a setup that had been quietly routing GitHub Copilot work to Anthropic’s Claude. Microsoft spent May telling its Windows, Office and Surface engineers to wind down Claude Code and move to GitHub Copilot CLI before the fiscal year closed on June […]<\/p>\n","protected":false},"author":1,"featured_media":9190,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[31],"tags":[],"class_list":["post-9189","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-31"],"magazineBlocksPostFeaturedMedia":{"thumbnail":"https:\/\/koshalsambada.in\/wp-content\/uploads\/2026\/08\/microsofts-new-internal-memo-makes-openais-gpt-56-the-default-model-for-its-engineers-replacing-a-se-150x150.avif","medium":"https:\/\/koshalsambada.in\/wp-content\/uploads\/2026\/08\/microsofts-new-internal-memo-makes-openais-gpt-56-the-default-model-for-its-engineers-replacing-a-se-300x169.avif","medium_large":"https:\/\/koshalsambada.in\/wp-content\/uploads\/2026\/08\/microsofts-new-internal-memo-makes-openais-gpt-56-the-default-model-for-its-engineers-replacing-a-se.avif","large":"https:\/\/koshalsambada.in\/wp-content\/uploads\/2026\/08\/microsofts-new-internal-memo-makes-openais-gpt-56-the-default-model-for-its-engineers-replacing-a-se.avif","1536x1536":"https:\/\/koshalsambada.in\/wp-content\/uploads\/2026\/08\/microsofts-new-internal-memo-makes-openais-gpt-56-the-default-model-for-its-engineers-replacing-a-se.avif","2048x2048":"https:\/\/koshalsambada.in\/wp-content\/uploads\/2026\/08\/microsofts-new-internal-memo-makes-openais-gpt-56-the-default-model-for-its-engineers-replacing-a-se.avif","blogsy-small":"https:\/\/koshalsambada.in\/wp-content\/uploads\/2026\/08\/microsofts-new-internal-memo-makes-openais-gpt-56-the-default-model-for-its-engineers-replacing-a-se.avif","blogsy-small-tall":"https:\/\/koshalsambada.in\/wp-content\/uploads\/2026\/08\/microsofts-new-internal-memo-makes-openais-gpt-56-the-default-model-for-its-engineers-replacing-a-se.avif","blogsy-small-square":"https:\/\/koshalsambada.in\/wp-content\/uploads\/2026\/08\/microsofts-new-internal-memo-makes-openais-gpt-56-the-default-model-for-its-engineers-replacing-a-se.avif","blogsy-small-masonry":"https:\/\/koshalsambada.in\/wp-content\/uploads\/2026\/08\/microsofts-new-internal-memo-makes-openais-gpt-56-the-default-model-for-its-engineers-replacing-a-se.avif","blogsy-medium":"https:\/\/koshalsambada.in\/wp-content\/uploads\/2026\/08\/microsofts-new-internal-memo-makes-openais-gpt-56-the-default-model-for-its-engineers-replacing-a-se.avif","blogsy-medium-masonry":"https:\/\/koshalsambada.in\/wp-content\/uploads\/2026\/08\/microsofts-new-internal-memo-makes-openais-gpt-56-the-default-model-for-its-engineers-replacing-a-se.avif","blogsy-large":"https:\/\/koshalsambada.in\/wp-content\/uploads\/2026\/08\/microsofts-new-internal-memo-makes-openais-gpt-56-the-default-model-for-its-engineers-replacing-a-se.avif","blogsy-wide":"https:\/\/koshalsambada.in\/wp-content\/uploads\/2026\/08\/microsofts-new-internal-memo-makes-openais-gpt-56-the-default-model-for-its-engineers-replacing-a-se.avif"},"magazineBlocksPostAuthor":{"name":"admin","avatar":"https:\/\/secure.gravatar.com\/avatar\/8709732a479614e7a8aa24d3eb1b239f30dc6d90c61464ed495001e7a469d856?s=96&d=mm&r=g"},"magazineBlocksPostCommentsNumber":"0","magazineBlocksPostExcerpt":"Microsoft’s new internal memo makes OpenAI’s GPT-5.6 the default model for its engineers, replacing a setup that had been quietly routing GitHub Copilot work to Anthropic’s Claude. Microsoft spent May telling its Windows, Office and Surface engineers to wind down Claude Code and move to GitHub Copilot CLI before the fiscal year closed on June […]","magazineBlocksPostCategories":["\u0b26\u0b47\u0b36 \u0b2c\u0b3f\u0b26\u0b47\u0b36"],"magazineBlocksPostViewCount":1,"magazineBlocksPostReadTime":3,"magazine_blocks_featured_image_url":{"full":["https:\/\/koshalsambada.in\/wp-content\/uploads\/2026\/08\/microsofts-new-internal-memo-makes-openais-gpt-56-the-default-model-for-its-engineers-replacing-a-se.avif",400,225,false],"medium":["https:\/\/koshalsambada.in\/wp-content\/uploads\/2026\/08\/microsofts-new-internal-memo-makes-openais-gpt-56-the-default-model-for-its-engineers-replacing-a-se-300x169.avif",300,169,true],"thumbnail":["https:\/\/koshalsambada.in\/wp-content\/uploads\/2026\/08\/microsofts-new-internal-memo-makes-openais-gpt-56-the-default-model-for-its-engineers-replacing-a-se-150x150.avif",150,150,true]},"magazine_blocks_author":{"display_name":"admin","author_link":"https:\/\/koshalsambada.in\/author\/admin"},"magazine_blocks_comment":0,"magazine_blocks_author_image":"https:\/\/secure.gravatar.com\/avatar\/8709732a479614e7a8aa24d3eb1b239f30dc6d90c61464ed495001e7a469d856?s=96&d=mm&r=g","magazine_blocks_category":"\u0b26\u0b47\u0b36 \u0b2c\u0b3f\u0b26\u0b47\u0b36<\/a>","_links":{"self":[{"href":"https:\/\/koshalsambada.in\/index.php?rest_route=\/wp\/v2\/posts\/9189","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/koshalsambada.in\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/koshalsambada.in\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/koshalsambada.in\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/koshalsambada.in\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=9189"}],"version-history":[{"count":0,"href":"https:\/\/koshalsambada.in\/index.php?rest_route=\/wp\/v2\/posts\/9189\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/koshalsambada.in\/index.php?rest_route=\/wp\/v2\/media\/9190"}],"wp:attachment":[{"href":"https:\/\/koshalsambada.in\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=9189"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/koshalsambada.in\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=9189"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/koshalsambada.in\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=9189"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}GitHub Copilot’s auto-router was quietly sending Microsoft engineers to Claude<\/b><\/h2>\n<\/p>\n
Microsoft’s AI token budget targets now put a ceiling on every division<\/b><\/h2>\n<\/p>\n
Anthropic keeps its Microsoft deals but loses the GitHub Copilot default<\/b><\/h2>\n<\/p>\n
\n
Source link <\/a><\/p>\n","protected":false},"excerpt":{"rendered":"