--- title: AT&T Shifts Most of Its AI Workload to Open Models description: AT&T is routing roughly 40% of employee AI usage to open-weight models instead of frontier models like those from OpenAI and Anthropic, with plans to raise... canonical: https://daily.dev/posts/at-t-shifts-most-of-its-ai-workload-to-open-models-rchjvssay twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: AT&T Shifts Most of Its AI Workload to Open Models | daily.dev og:description: AT&T is routing roughly 40% of employee AI usage to open-weight models instead of frontier models like those from OpenAI and Anthropic, with plans to raise... og:url: https://daily.dev/posts/at-t-shifts-most-of-its-ai-workload-to-open-models-rchjvssay og:image: https://api.daily.dev/og/posts/rCHjvSSAY.png og:image:alt: AT&T Shifts Most of Its AI Workload to Open Models 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. # AT&T Shifts Most of Its AI Workload to Open Models **[Collections](https://daily.dev/sources/collections)** · 2 min read · 0 upvotes · 0 comments ## Summary AT&T is routing roughly 40% of employee AI usage to open-weight models instead of frontier models like those from OpenAI and Anthropic, with plans to raise that share to 60-70%. The company reports coding costs down 56% while output quality dropped only 2%, and it processes about 45 billion tokens per day. Frontier models are still reserved for high-stakes, critical work. The move is framed as a signal that enterprises may increasingly resist paying premium prices for frontier models when open alternatives deliver comparable quality at much lower cost, echoing spending trends showing price acting as a ceiling on frontier model adoption. ## Content AT&T just gave enterprise AI vendors a preview of their worst nightmare: a large customer quietly routing the bulk of its work to open models instead of paying premium prices for frontier ones. According to AT&T's AI chief, the company now sends about 40% of employee AI usage to open models, with plans to push that to 60-70%. The numbers behind the move are hard to argue with: - Coding costs down 56% - Output quality down only 2% - 45 billion tokens processed per day AT&T isn't abandoning frontier models entirely. Anthropic and OpenAI's systems still handle the critical, high-stakes work. But for everything else, AT&T's own AI chief says open models are "just as good or better" for many tasks. That's a pretty blunt way to describe the middle of your product line as expendable. This lines up with what Ramp's spending data has been showing: Anthropic has been pulling further ahead of OpenAI in enterprise, but price is clearly acting as a ceiling on how far that lead can go. When a customer the size of AT&T can cut costs in half with a 2% quality tradeoff, that's not a rounding error - that's a signal other enterprises are going to notice. I keep coming back to that 2% number. If that gap holds up across other companies' workloads, it's hard to see why more of them wouldn't make the same trade. This looks less like an AT&T story and more like the first crack in the assumption that enterprises will keep paying a premium for frontier models no matter what. ## Questions this post answers ### How much has AT&T saved by switching from frontier AI models to open models for coding tasks? AT&T cut coding costs by 56% by routing that work to open models instead of frontier models like those from OpenAI and Anthropic, while output quality dropped only about 2%. The company currently sends roughly 40% of employee AI usage to open models, processing about 45 billion tokens per day, and plans to increase that share to 60-70%. _Enterprises weighing frontier versus open model costs can follow real-world adoption shifts like this on daily.dev._ ### Are companies still using frontier models like GPT and Claude for anything if open models are cheaper? Yes, AT&T still relies on frontier models from Anthropic and OpenAI for critical, high-stakes work, while shifting routine tasks to open models where its AI chief says they perform just as well or better for many use cases. The split reflects a cost-quality tradeoff rather than a full abandonment of frontier providers. _Track how enterprises split workloads between frontier and open models on daily.dev._ ## Community take How the wider developer community reacted, aggregated from 2 discussions and 83 comments across x (as of 2026-08-21). **TL;DR:** Reactions are split between enthusiasm that a major enterprise is proving open-weight models can cut costs dramatically with minimal quality loss, and skepticism about whether the 2% quality claim, the token volume, or the whole framing is meaningful or trustworthy. **Sentiment:** 30% positive · 45% mixed · 25% skeptical **The case for** - Some see this as clean proof that open models are good enough for the bulk of enterprise workloads, freeing frontier models for only the hardest tasks. - Several argue this signals a broader shift where routing pipelines make switching away from frontier labs cheap and easy, threatening their pricing power. - A few note real-world personal experience already doing the same thing (e.g., using Qwen instead of Claude for PDF extraction) at much lower cost. **The pushback** - Multiple commenters doubt the '2% quality drop' figure, arguing a doubling of output can't realistically cost so little in quality and question which eval was used. - Some think the 45B tokens/day figure is either unbelievable, trivial in dollar terms (~$35K/day), or a sign of inefficient architecture rather than genuine capability. - Others argue this is just tiering like cloud commoditization, not a real threat to frontier labs, since hard work still goes upstream to frontier models. **By community** - x (mixed): Replies mix genuine enthusiasm about the cost/quality tradeoff with pointed skepticism about the reported numbers and what they actually prove. **Hottest debate:** Whether the reported 2% quality drop and 45B token figure are meaningful signals of open models catching up, or misleading/inflated numbers that don't hold up to scrutiny. **Open questions** - What benchmark or eval was used to measure the '2% quality drop'? - What exactly counts as 'critical' work still routed to frontier models versus everything else? - Which inference provider or infrastructure is AT&T actually using to run these open models? **Highlights** > @Hesamation > quality dropped 2% Yeah, no. If your output is doubled your quality goes down more than that. That's the trouble with people who have no idea what software engineers do reporting on shit like this > — [thx6969 on x · 9 points, 4 comments](https://x.com/thx6969/status/2090584919924003085) > @Hesamation You do realize 45B tokens / day is something like $35K in spend yes? Legit meaningless. > — [drapersgulld on x · 2 points](https://x.com/drapersgulld/status/2090592958361841759) > @Hesamation They won't get better results with open source. They're burning 45B tokens a day because they don't have a f'ing clue how to build efficient architecture. Management yells "Go find alpha!" but the guys at AT&T were never going to find it anyway. So much BS being spewed. > — [LibertyLynx on x · 1 comments](https://x.com/LibertyLynx/status/2090643006118567957) > @Hesamation Read the pull quote, not the headline: open models beat older frontier models, and AT&T still routes the hard work upstream. That's not disintermediation, it's tiering — the same thing that happened to cloud, where commodity absorbed the volume and the premium tier kept the > — [ricci\_nov on x](https://x.com/ricci_nov/status/2090630518635155776) > @Hesamation quality down 2%, coding costs down 56%, and AT&T is taking open models from 40% to 70% of internal usage. the 2% is self-reported, the 56% is the number every enterprise CFO will remember. 45b tokens a day is the real headline. > — [first\_sauce\_lab on x](https://x.com/first_sauce_lab/status/2090594754618618306) **Source threads** - [x](https://x.com/mervenoyann/status/2090573563048730659) · 0 points · 0 comments - [x](https://x.com/Hesamation/status/2090518831349268851) · 0 points · 83 comments ## Similar posts on daily.dev - [Mistral Says 99% of Enterprise AI Can Be Open. That Should Terrify the Closed-Model Business](https://daily.dev/posts/mistral-says-99-of-enterprise-ai-can-be-open-that-should-terrify-the-closed-model-business-eq4tuzn9x) · Medium · 1 upvotes · 0 comments --- Tags: [#anthropic](https://daily.dev/tags/anthropic), [#llm](https://daily.dev/tags/llm), [#openai](https://daily.dev/tags/openai), [#open-source](https://daily.dev/tags/open-source) [View this post on daily.dev](https://daily.dev/posts/at-t-shifts-most-of-its-ai-workload-to-open-models-rchjvssay) ```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":"AT&T Shifts Most of Its AI Workload to Open Models","url":"https://daily.dev/posts/at-t-shifts-most-of-its-ai-workload-to-open-models-rchjvssay","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/at-t-shifts-most-of-its-ai-workload-to-open-models-rchjvssay"},"datePublished":"2026-08-20T22:56:24.536Z","dateModified":"2026-10-15T02:14:46.149Z","description":"AT&T is routing roughly 40% of employee AI usage to open-weight models instead of frontier models like those from OpenAI and Anthropic, with plans to raise...","isAccessibleForFree":true,"articleSection":"Collections","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":"Collections","logo":"https://media.daily.dev/image/upload/s--fk_6ycEi--/f_auto,q_auto/v1780996001/logos/collections?_a=BAMAMiWQ0","url":"https://daily.dev/sources/collections"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/at-t-shifts-most-of-its-ai-workload-to-open-models-rchjvssay","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":0},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"anthropic,llm,openai,open-source","timeRequired":"PT2M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Collections","item":"https://daily.dev/sources/collections"},{"@type":"ListItem","position":3,"name":"AT&T Shifts Most of Its AI Workload to Open Models"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/at-t-shifts-most-of-its-ai-workload-to-open-models-rchjvssay#faq","mainEntity":[{"@type":"Question","name":"How much has AT&T saved by switching from frontier AI models to open models for coding tasks?","acceptedAnswer":{"@type":"Answer","text":"AT&T cut coding costs by 56% by routing that work to open models instead of frontier models like those from OpenAI and Anthropic, while output quality dropped only about 2%. The company currently sends roughly 40% of employee AI usage to open models, processing about 45 billion tokens per day, and plans to increase that share to 60-70%. Enterprises weighing frontier versus open model costs can follow real-world adoption shifts like this on daily.dev."}},{"@type":"Question","name":"Are companies still using frontier models like GPT and Claude for anything if open models are cheaper?","acceptedAnswer":{"@type":"Answer","text":"Yes, AT&T still relies on frontier models from Anthropic and OpenAI for critical, high-stakes work, while shifting routine tasks to open models where its AI chief says they perform just as well or better for many use cases. The split reflects a cost-quality tradeoff rather than a full abandonment of frontier providers. Track how enterprises split workloads between frontier and open models on daily.dev."}}]} ```