--- title: This is absolute chaos... | daily.dev description: GPT-4.1 (referred to as o3/o4 'Soul' in context) burns through rate limits far faster than its predecessor due to its more autonomous, non-stopping behavior.... canonical: https://daily.dev/posts/this-is-absolute-chaos--jdlpenmxh twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: This is absolute chaos... | daily.dev og:description: GPT-4.1 (referred to as o3/o4 'Soul' in context) burns through rate limits far faster than its predecessor due to its more autonomous, non-stopping behavior.... og:url: https://daily.dev/posts/this-is-absolute-chaos--jdlpenmxh og:image: https://api.daily.dev/og/posts/jdlpeNMXh.png og:image:alt: This is absolute chaos... 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. # This is absolute chaos... **[Theo - t3․gg](https://daily.dev/sources/t3dotgg)** · 30 min read · 1 upvotes · 0 comments ## Summary GPT-4.1 (referred to as o3/o4 'Soul' in context) burns through rate limits far faster than its predecessor due to its more autonomous, non-stopping behavior. Key tips to reduce usage: disable Fast mode (2.5x token burn for minimal speed gain), avoid Ultra mode entirely, stick to 'High' reasoning level as the sweet spot between cost and performance, restrict sub-agent usage via agents.md unless explicitly requested, and use explicit stop points in prompts to control how far the model runs. OpenAI has temporarily removed 5-hour limits (weekly limit now applies), rolled back a context window expansion that caused overcharging, and is shipping inference optimizations. The OpenCode team accidentally ran on Medium reasoning for a month and still found it their favorite model, endorsing Medium or High as solid defaults. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### Why am I suddenly hitting Codex usage limits so much faster with GPT-5.6 Soul than I did with GPT-5.5? GPT-5.6 Soul fixed GPT-5.5's tendency to stop and ask for permission mid-task, so it now runs much longer per message, using up to 15% of a 5-hour limit in one message versus 0.1-2% for 5.5. Combined with fast mode's 2.5x usage multiplier, a single high-reasoning message can consume nearly half a 5-hour limit. _Developers tuning Codex reasoning levels to avoid rate limits can track these kinds of usage shifts on daily.dev._ ### Should I turn on fast mode in OpenAI Codex to speed up coding tasks? No, fast mode is not recommended with GPT-5.6 Soul because it gives only 1.5x faster inference but burns through usage limits 2.5x faster, and since the model now runs long tasks autonomously rather than stopping for feedback, the speed benefit is negligible compared to the usage cost. _Anyone optimizing AI coding workflows for cost versus speed can follow updates like this on daily.dev._ ### What did OpenAI temporarily change about Codex usage limits after GPT-5.6 Soul launched? OpenAI temporarily removed the 5-hour usage limits for Plus, Business, and Pro plans, leaving only the weekly limit active, and shipped a 10% inference cost reduction. They also found that raising the context window to 372K from 272K caused unintended extra usage charges due to a tracking bug, so they reverted it back to 272K while a proper fix is prepared. _Teams relying on Codex subscription limits for daily work can watch for changes like this on daily.dev._ --- Tags: [#ai-coding](https://daily.dev/tags/ai-coding), [#prompt-engineering](https://daily.dev/tags/prompt-engineering), [#gpt](https://daily.dev/tags/gpt), [#openai-codex](https://daily.dev/tags/openai-codex) [View this post on daily.dev](https://daily.dev/posts/this-is-absolute-chaos--jdlpenmxh) ```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":"This is absolute chaos...","url":"https://daily.dev/posts/this-is-absolute-chaos--jdlpenmxh","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/this-is-absolute-chaos--jdlpenmxh"},"datePublished":"2026-07-13T08:28:20.265Z","dateModified":"2026-09-14T09:08:44.596Z","description":"GPT-4.1 (referred to as o3/o4 'Soul' in context) burns through rate limits far faster than its predecessor due to its more autonomous, non-stopping behavior....","image":"https://i.ytimg.com/vi/sKmrLtB47WA/sddefault.jpg","thumbnailUrl":"https://i.ytimg.com/vi/sKmrLtB47WA/sddefault.jpg","isAccessibleForFree":true,"articleSection":"Theo - t3․gg","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":"Theo - t3․gg","logo":"https://media.daily.dev/image/upload/s--UmX7IyU3--/f_auto/v1704628081/logos/t3dotgg.jpg","url":"https://daily.dev/sources/t3dotgg"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/this-is-absolute-chaos--jdlpenmxh","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":1},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"ai-coding,prompt-engineering,gpt,openai-codex","timeRequired":"PT30M","video":{"@type":"VideoObject","name":"This is absolute chaos...","description":"GPT-4.1 (referred to as o3/o4 'Soul' in context) burns through rate limits far faster than its predecessor due to its more autonomous, non-stopping behavior....","thumbnailUrl":"https://i.ytimg.com/vi/sKmrLtB47WA/sddefault.jpg","uploadDate":"2026-07-13T08:28:20.265Z","duration":"PT30M","url":"https://api.daily.dev/r/jdlpeNMXh","embedUrl":"https://www.youtube.com/embed/sKmrLtB47WA"}} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Theo - t3․gg","item":"https://daily.dev/sources/t3dotgg"},{"@type":"ListItem","position":3,"name":"This is absolute chaos..."}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/this-is-absolute-chaos--jdlpenmxh#faq","mainEntity":[{"@type":"Question","name":"Why am I suddenly hitting Codex usage limits so much faster with GPT-5.6 Soul than I did with GPT-5.5?","acceptedAnswer":{"@type":"Answer","text":"GPT-5.6 Soul fixed GPT-5.5's tendency to stop and ask for permission mid-task, so it now runs much longer per message, using up to 15% of a 5-hour limit in one message versus 0.1-2% for 5.5. Combined with fast mode's 2.5x usage multiplier, a single high-reasoning message can consume nearly half a 5-hour limit. Developers tuning Codex reasoning levels to avoid rate limits can track these kinds of usage shifts on daily.dev."}},{"@type":"Question","name":"Should I turn on fast mode in OpenAI Codex to speed up coding tasks?","acceptedAnswer":{"@type":"Answer","text":"No, fast mode is not recommended with GPT-5.6 Soul because it gives only 1.5x faster inference but burns through usage limits 2.5x faster, and since the model now runs long tasks autonomously rather than stopping for feedback, the speed benefit is negligible compared to the usage cost. Anyone optimizing AI coding workflows for cost versus speed can follow updates like this on daily.dev."}},{"@type":"Question","name":"What did OpenAI temporarily change about Codex usage limits after GPT-5.6 Soul launched?","acceptedAnswer":{"@type":"Answer","text":"OpenAI temporarily removed the 5-hour usage limits for Plus, Business, and Pro plans, leaving only the weekly limit active, and shipped a 10% inference cost reduction. They also found that raising the context window to 372K from 272K caused unintended extra usage charges due to a tracking bug, so they reverted it back to 272K while a proper fix is prepared. Teams relying on Codex subscription limits for daily work can watch for changes like this on daily.dev."}}]} ```