--- title: Open models hit 62% of token usage, and nobody saw it coming description: Open-weight AI models jumped from 28.4% to 62% of token traffic on Vercel's AI Gateway between June and August 22, according to Vercel CEO Guillermo Rauch. He... canonical: https://daily.dev/posts/open-models-hit-62-of-token-usage-and-nobody-saw-it-coming-kr9tmxeor twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: Open models hit 62% of token usage, and nobody saw it coming | daily.dev og:description: Open-weight AI models jumped from 28.4% to 62% of token traffic on Vercel's AI Gateway between June and August 22, according to Vercel CEO Guillermo Rauch. He... og:url: https://daily.dev/posts/open-models-hit-62-of-token-usage-and-nobody-saw-it-coming-kr9tmxeor og:image: https://api.daily.dev/og/posts/Kr9TmxEor.png og:image:alt: Open models hit 62% of token usage, and nobody saw it coming 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. # Open models hit 62% of token usage, and nobody saw it coming **[Trends](https://daily.dev/sources/trends)** · 2 min read · 5 upvotes · 0 comments ## Summary Open-weight AI models jumped from 28.4% to 62% of token traffic on Vercel's AI Gateway between June and August 22, according to Vercel CEO Guillermo Rauch. He argues this is early-stage momentum since most developer tooling (CLIs, IDEs, SDKs) is still hardcoded to closed models, and once tooling becomes model-agnostic, adoption could accelerate further. Enterprises are driving the shift for cost reasons: AT&T reportedly cut inference costs by 56% and Coinbase by around 50% by routing to open models. A SemiAnalysis chart shows the quality gap between open and closed frontier models has narrowed to about 3.3% on Stanford's benchmarks, or roughly four months per Epoch AI estimates. ## Content The number that's driving the conversation: open-weight models went from 28.4% to 62% of tokens on Vercel's AI Gateway between June and August. Two months. That's not a trend line, that's a cliff. Vercel CEO Guillermo Rauch called August 22 a record day and noted this is "very likely just the start" because enterprise adoption is still early and most tooling isn't model-agnostic yet. That last part matters — the 62% figure is happening *before* the infrastructure fully catches up. Chamath put the pricing reality bluntly: "A barrel of intelligence" (his term for one million tokens) costs $26 from OpenAI, $56 from Anthropic, $1 from xAI, $1.50 from Meta, and $0.50 from Chinese providers. "If you made an early bet on one of the companies selling extremely expensive barrels of intelligence, and you try to pass those costs through to customers, you may run into downstream difficulties." Enterprises are already acting on this. AT&T cut costs 56% by routing to cheaper models. Coinbase saved roughly 50%. Cisco built an agent system for 90,000 employees where 50-60% of requests go to open-weight models, with only a small slice hitting frontier closed models — and that's by design, not accident. The multi-agent pattern is accelerating this. As one ML researcher pointed out, in supervisor-plus-subagent architectures, the subagents handle most of the token volume. Frontier open models are "beautifully" suited for that role — cheaper, fast enough, and good enough. There's no reason to run Claude Opus on every subagent call when a smaller model handles it fine. The quality gap is also closing faster than expected. Six months ago, Claude Opus 4.6 was state of the art. Today, Qwen 3.8 — a 27B model that runs on a gaming GPU — beats it on some coding benchmarks. The gap Stanford measured at 3.3%, Epoch AI at roughly four months of lag. One real caveat: "open" is getting complicated. Kimi K3 and MiniMax M3 require revenue-based licensing agreements. DeepSeek V4 and GLM 5.2 use permissive MIT licenses. Downloadable weights don't automatically mean unrestricted commercial use. Enterprises building on this shift should read the licenses before assuming they're in the clear. But the direction is set. The token market has already flipped. ## Questions this post answers ### What percentage of tokens on Vercel's AI Gateway are now going to open-weight models instead of closed frontier models? Open-weight models grew from 28.4% to 62% of token usage on Vercel's AI Gateway between June and August, according to Vercel CEO Guillermo Rauch, who called August 22 a record day. He expects the shift to continue since enterprise adoption is still early and most tooling isn't yet model-agnostic. _Track how fast open-weight adoption is reshaping model costs and vendor choices on daily.dev._ ### How much cheaper are open-weight LLM tokens compared to OpenAI and Anthropic models? One million tokens costs roughly $26 from OpenAI and $56 from Anthropic, compared to about $1 from xAI, $1.50 from Meta, and $0.50 from Chinese providers, according to figures cited by Chamath. Enterprises betting on the pricier closed models risk difficulty passing those costs through to customers as cheaper alternatives close the quality gap. _Compare token pricing across providers on daily.dev before locking a team into one model vendor._ ### Are all open-weight AI models free to use commercially? No, downloadable weights don't automatically mean unrestricted commercial use. Kimi K3 and MiniMax M3 require revenue-based licensing agreements, while DeepSeek V4 and GLM 5.2 use permissive MIT licenses, so enterprises should read licensing terms before assuming they can deploy a model freely. _Check model licensing details on daily.dev before building production systems on open-weight models._ ## Community take How the wider developer community reacted, aggregated from 1 discussion and 1 comments across x (as of 2026-09-13). **TL;DR:** The single available reply frames the shift to open-weight models as a straightforward cost-driven business decision, not a surprise. **Sentiment:** 40% positive · 20% mixed · 40% skeptical **The case for** - Adopting cheaper open-weight models is a rational cost-saving strategy for businesses. **By community** - x (mixed): A lone reply argues low cost is the deliberate strategy behind adoption, not an unexpected trend. **Highlights** > @trikcode Low cost is the whole point though US businesses adopting because its cheap isnt a surprise, its the strategy > — [HarryTandy on x](https://x.com/HarryTandy/status/2093373747306672265) **Source threads** - [x](https://x.com/trikcode/status/2093372305887862895) · 0 points · 1 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 - [The State of Open Source AI — V1.0 · July 2026](https://daily.dev/posts/the-state-of-open-source-ai-v1-0-july-2026-gjzwgazbw) · Hacker News · 1 upvotes · 0 comments --- Tags: [#vercel](https://daily.dev/tags/vercel), [#finops](https://daily.dev/tags/finops), [#ai-inference](https://daily.dev/tags/ai-inference), [#ai-gateway](https://daily.dev/tags/ai-gateway) [View this post on daily.dev](https://daily.dev/posts/open-models-hit-62-of-token-usage-and-nobody-saw-it-coming-kr9tmxeor) ```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":"Open models hit 62% of token usage, and nobody saw it coming","url":"https://daily.dev/posts/open-models-hit-62-of-token-usage-and-nobody-saw-it-coming-kr9tmxeor","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/open-models-hit-62-of-token-usage-and-nobody-saw-it-coming-kr9tmxeor"},"datePublished":"2026-08-23T00:05:35.991Z","dateModified":"2026-09-13T19:50:12.902Z","description":"Open-weight AI models jumped from 28.4% to 62% of token traffic on Vercel's AI Gateway between June and August 22, according to Vercel CEO Guillermo Rauch. He...","image":"https://pbs.twimg.com/media/HQXYk9eWgAAbVqe.jpg","thumbnailUrl":"https://pbs.twimg.com/media/HQXYk9eWgAAbVqe.jpg","isAccessibleForFree":true,"articleSection":"Trends","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":"Trends","logo":"https://media.daily.dev/image/upload/s--ZfSp3asX--/f_auto,q_auto/v1780996004/logos/trends?_a=BAMAMiWQ0","url":"https://daily.dev/sources/trends"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/open-models-hit-62-of-token-usage-and-nobody-saw-it-coming-kr9tmxeor","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":5},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"vercel,finops,ai-inference,ai-gateway","timeRequired":"PT2M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Trends","item":"https://daily.dev/sources/trends"},{"@type":"ListItem","position":3,"name":"Open models hit 62% of token usage, and nobody saw it coming"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/open-models-hit-62-of-token-usage-and-nobody-saw-it-coming-kr9tmxeor#faq","mainEntity":[{"@type":"Question","name":"What percentage of tokens on Vercel's AI Gateway are now going to open-weight models instead of closed frontier models?","acceptedAnswer":{"@type":"Answer","text":"Open-weight models grew from 28.4% to 62% of token usage on Vercel's AI Gateway between June and August, according to Vercel CEO Guillermo Rauch, who called August 22 a record day. He expects the shift to continue since enterprise adoption is still early and most tooling isn't yet model-agnostic. Track how fast open-weight adoption is reshaping model costs and vendor choices on daily.dev."}},{"@type":"Question","name":"How much cheaper are open-weight LLM tokens compared to OpenAI and Anthropic models?","acceptedAnswer":{"@type":"Answer","text":"One million tokens costs roughly $26 from OpenAI and $56 from Anthropic, compared to about $1 from xAI, $1.50 from Meta, and $0.50 from Chinese providers, according to figures cited by Chamath. Enterprises betting on the pricier closed models risk difficulty passing those costs through to customers as cheaper alternatives close the quality gap. Compare token pricing across providers on daily.dev before locking a team into one model vendor."}},{"@type":"Question","name":"Are all open-weight AI models free to use commercially?","acceptedAnswer":{"@type":"Answer","text":"No, downloadable weights don't automatically mean unrestricted commercial use. Kimi K3 and MiniMax M3 require revenue-based licensing agreements, while DeepSeek V4 and GLM 5.2 use permissive MIT licenses, so enterprises should read licensing terms before assuming they can deploy a model freely. Check model licensing details on daily.dev before building production systems on open-weight models."}}]} ```