{"id":2630,"date":"2026-04-24T09:26:41","date_gmt":"2026-04-24T09:26:41","guid":{"rendered":"https:\/\/deepinsightai.io\/?p=2630"},"modified":"2026-04-25T06:30:31","modified_gmt":"2026-04-25T06:30:31","slug":"claude-admits-getting-dumber","status":"publish","type":"post","link":"https:\/\/deepinsightai.io\/fr\/claude-admits-getting-dumber\/","title":{"rendered":"Claude Admits \u201cGetting Dumber\u201d: 3 Critical Bugs Exposed After GPT-5.5 Launch"},"content":{"rendered":"<p>Right around the release of <a href=\"https:\/\/deepinsightai.io\/fr\/gpt-5-5-review\/\" target=\"_blank\" rel=\"noreferrer noopener\">GPT-5.5<\/a>, Claude finally admitted it:<\/p>\n\n\n\n<figure data-spectra-id=\"spectra-mocpcmrs-1ng537\" class=\"wp-block-image aligncenter size-full\"><img fetchpriority=\"high\" decoding=\"async\" width=\"570\" height=\"135\" src=\"https:\/\/deepinsightai.io\/wp-content\/uploads\/2026\/04\/image-78.png\" alt=\"Claude Admits \u201cGetting Dumber\" class=\"wp-image-2633\" title=\"Claude Admits \u201cGetting Dumber\u201d: 3 Critical Bugs Exposed After GPT-5.5 Launch\" srcset=\"https:\/\/deepinsightai.io\/wp-content\/uploads\/2026\/04\/image-78.png 570w, https:\/\/deepinsightai.io\/wp-content\/uploads\/2026\/04\/image-78-300x71.png 300w, https:\/\/deepinsightai.io\/wp-content\/uploads\/2026\/04\/image-78-18x4.png 18w\" sizes=\"(max-width: 570px) 100vw, 570px\" \/><\/figure>\n\n\n\n<p>Yes, the model got \u201cdumber,\u201d and all usage quotas have been reset.<\/p>\n\n\n\n<p>After over a month of denial, the intelligence drop bug finally came straight from Anthropic itself:<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>Reasoning level quietly downgraded from \u201chigh\u201d to \u201cmedium\u201d<\/li>\n\n\n\n<li>A caching bug wiped thinking traces every turn<\/li>\n\n\n\n<li>A 25-word prompt limit crushed output quality<\/li>\n<\/ul>\n\n\n\n<p>Stack those three bugs together, and Claude\u2019s overall experience took a serious hit.<\/p>\n\n\n\n<p>Good thing competitors applied pressure at just the right moment. Trying to \u201ceducate users\u201d clearly wasn\u2019t going to work.<\/p>\n\n\n\n<p>Still, fixing bugs is always a good thing. It\u2019s just\u2026 the timing is suspicious. GPT-5.5 drops, and suddenly Claude \u201cadmits defeat.\u201d<\/p>\n\n\n\n<p>Was 5.5 helping debug?<\/p>\n\n\n\n<p>Dario\u2026 you didn\u2019t intentionally make Claude worse just to set up a comeback moment when GPT-5.5 launched, did you?<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">Three confirmed bugs behind Claude\u2019s performance drop<\/h2>\n\n\n\n<p>First, this isn\u2019t the first time.<\/p>\n\n\n\n<p>Back in August last year, Anthropic published a similar postmortem affecting Opus 4.0 and 4.1, also stating \u201cwe never intended to reduce model quality.\u201d<\/p>\n\n\n\n<p>This time, the new postmortem is titled <em>\u201cA postmortem of three recent issues.\u201d<\/em><\/p>\n\n\n\n<p>\u201cRecent\u201d is key \u2014 not just now, but over the past period.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">The community had already noticed<\/h3>\n\n\n\n<p>Claude getting worse had been widely discussed for a while.<\/p>\n\n\n\n<p>Over ten days ago, AMD AI senior director Stella Laurenzo published a deep audit on GitHub:<\/p>\n\n\n\n<figure data-spectra-id=\"spectra-mocpdaju-r03hld\" class=\"wp-block-image aligncenter size-full\"><img decoding=\"async\" width=\"681\" height=\"345\" src=\"https:\/\/deepinsightai.io\/wp-content\/uploads\/2026\/04\/image-79.png\" alt=\"AMD AI senior director Stella Laurenzo published a deep audit on GitHub\" class=\"wp-image-2634\" title=\"Claude Admits \u201cGetting Dumber\u201d: 3 Critical Bugs Exposed After GPT-5.5 Launch\" srcset=\"https:\/\/deepinsightai.io\/wp-content\/uploads\/2026\/04\/image-79.png 681w, https:\/\/deepinsightai.io\/wp-content\/uploads\/2026\/04\/image-79-300x152.png 300w, https:\/\/deepinsightai.io\/wp-content\/uploads\/2026\/04\/image-79-18x9.png 18w\" sizes=\"(max-width: 681px) 100vw, 681px\" \/><\/figure>\n\n\n\n<ul class=\"wp-block-list\">\n<li>6,852 conversations<\/li>\n\n\n\n<li>17,871 reasoning blocks<\/li>\n\n\n\n<li>230,000+ tool calls<\/li>\n<\/ul>\n\n\n\n<p>The findings: since February, reasoning depth dropped sharply.<\/p>\n\n\n\n<p>More detailed observations:<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>Claude started falling into <strong>reasoning loops<\/strong><\/li>\n\n\n\n<li>It increasingly chose the <strong>simplest fix<\/strong>, not the correct one<\/li>\n<\/ul>\n\n\n\n<p>Meanwhile, BridgeMind\u2019s BridgeBench benchmark showed <a href=\"https:\/\/deepinsightai.io\/fr\/claude-opus-4-7-vs-opus-4-6\/\" target=\"_blank\" rel=\"noreferrer noopener\">Opus 4.6<\/a> accuracy falling from 83.3% to 68.3%, ranking from #2 to #10.<\/p>\n\n\n\n<figure data-spectra-id=\"spectra-mocpe4qx-d4yl2z\" class=\"wp-block-image aligncenter size-full\"><img decoding=\"async\" width=\"590\" height=\"830\" src=\"https:\/\/deepinsightai.io\/wp-content\/uploads\/2026\/04\/image-80.png\" alt=\"BridgeMind\u2019s BridgeBench benchmark showed Opus 4.6 accuracy falling from 83.3% to 68.3%, ranking from #2 to #10.\" class=\"wp-image-2635\" title=\"Claude Admits \u201cGetting Dumber\u201d: 3 Critical Bugs Exposed After GPT-5.5 Launch\" srcset=\"https:\/\/deepinsightai.io\/wp-content\/uploads\/2026\/04\/image-80.png 590w, https:\/\/deepinsightai.io\/wp-content\/uploads\/2026\/04\/image-80-213x300.png 213w, https:\/\/deepinsightai.io\/wp-content\/uploads\/2026\/04\/image-80-9x12.png 9w\" sizes=\"(max-width: 590px) 100vw, 590px\" \/><\/figure>\n\n\n\n<p>Even though the methodology was later criticized, the narrative stuck:<\/p>\n\n\n\n<p>Claude is getting dumber.<\/p>\n\n\n\n<p>People even coined a term \u2014 <strong>AI shrinkflation<\/strong>. Same price, diluted product.<\/p>\n\n\n\n<p>In other words, users felt like they were getting a \u201ckids\u2019 meal.\u201d<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">The three bugs explained<\/h2>\n\n\n\n<h3 class=\"wp-block-heading\">Silent downgrade of reasoning level<\/h3>\n\n\n\n<p>On March 4, Claude Code\u2019s default <a href=\"https:\/\/deepinsightai.io\/fr\/claude-opus-4-7-adaptive-thinking\/\" target=\"_blank\" rel=\"noreferrer noopener\">reasoning level<\/a> was lowered from \u201chigh\u201d to \u201cmedium\u201d due to latency concerns.<\/p>\n\n\n\n<p>But the UI still showed \u201chigh.\u201d<\/p>\n\n\n\n<p>Users thought they were using full power \u2014 but actually got a downgraded version.<\/p>\n\n\n\n<p>This lasted over a month before being rolled back.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">Gets dumber the longer you chat<\/h3>\n\n\n\n<p>On March 26, a caching optimization was introduced:<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>Intended: clear old reasoning after 1 hour of inactivity<\/li>\n\n\n\n<li>Actual: due to a bug, it cleared <strong>every single turn<\/strong><\/li>\n<\/ul>\n\n\n\n<p>R\u00e9sultat :<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>Claude kept working but forgot its own goals<\/li>\n\n\n\n<li>Symptoms: forgetfulness, repetition, chaotic tool use<\/li>\n<\/ul>\n\n\n\n<p>Worse:<\/p>\n\n\n\n<p>Because reasoning was constantly wiped, cache never hit \u2192 token usage skyrocketed.<\/p>\n\n\n\n<p>It took 15 days to fix.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">One prompt line crushed output quality<\/h3>\n\n\n\n<p>On April 16, a system instruction was added:<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>Max 25 words between tool calls<\/li>\n\n\n\n<li>Final reply limited to 100 words<\/li>\n<\/ul>\n\n\n\n<p>Impact:<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><a href=\"https:\/\/deepinsightai.io\/fr\/claude-opus-4-7\/\" target=\"_blank\" rel=\"noreferrer noopener\">Opus 4.6 and 4.7<\/a> both dropped ~3% performance<\/li>\n\n\n\n<li>Rolled back after 4 days<\/li>\n<\/ul>\n\n\n\n<h2 class=\"wp-block-heading\">April: a Series of Self-Inflicted Issues<\/h2>\n\n\n\n<p>Looking at the timeline, April wasn\u2019t just bugs \u2014 it was a series of questionable moves.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">April 4 \u2014 Blocking third-party tools<\/h3>\n\n\n\n<p>Anthropic blocked tools like <a href=\"https:\/\/deepinsightai.io\/fr\/openclaw-collapse\/\" target=\"_blank\" rel=\"noreferrer noopener\">OpenClaw<\/a> from using Pro\/Max subscriptions.<\/p>\n\n\n\n<p>Want to keep using them? Switch to API billing.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">April 21 \u2014 Quiet pricing changes<\/h3>\n\n\n\n<p><a href=\"https:\/\/deepinsightai.io\/fr\/did-anthropic-remove-claude-code-from-pro\/\" target=\"_blank\" rel=\"noreferrer noopener\">Claude Code was removed<\/a> from the Pro plan page.<\/p>\n\n\n\n<p>Support docs changed from:<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u201cPro or Max\u201d \u2192 \u201cMax only\u201d<\/li>\n<\/ul>\n\n\n\n<p>After backlash, Head of Growth Amol Avasare claimed it was just a 2% A\/B test.<\/p>\n\n\n\n<p>Problem: the change was site-wide.<\/p>\n\n\n\n<p>It got rolled back within hours.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">Pricing reality<\/h3>\n\n\n\n<ul class=\"wp-block-list\">\n<li>Pro: $20\/month \u2192 $240\/year<\/li>\n\n\n\n<li>Max 5x: $100\/month \u2192 $1200\/year (5\u00d7 increase)<\/li>\n\n\n\n<li>Max 20x: $2400\/year (10\u00d7 increase)<\/li>\n<\/ul>\n\n\n\n<p>No middle tier.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">April 23 \u2014 Postmortem and \u201ccompensation\u201d<\/h3>\n\n\n\n<p>The compensation? Reset usage quotas.<\/p>\n\n\n\n<p>Some users pointed out:<\/p>\n\n\n\n<p>That had already happened during the Opus 4.7 release.<\/p>\n\n\n\n<p>So this \u201ccompensation\u201d was basically just a normal cycle reset.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">Users aren\u2019t buying it<\/h2>\n\n\n\n<p>Reactions are split.<\/p>\n\n\n\n<p>Some say:<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>Bugs happen<\/li>\n\n\n\n<li>The postmortem is transparent<\/li>\n\n\n\n<li>Boris Cherny actively responded on Hacker News<\/li>\n<\/ul>\n\n\n\n<p>Others focus on a different issue:<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">Silence for two months<\/h3>\n\n\n\n<ul class=\"wp-block-list\">\n<li>No official communication<\/li>\n\n\n\n<li>Only scattered replies on X<\/li>\n\n\n\n<li>No consistent messaging<\/li>\n<\/ul>\n\n\n\n<h3 class=\"wp-block-heading\">Suspicion about \u201ccache optimization\u201d<\/h3>\n\n\n\n<p>Some believe the timing (aligned with cache expiration) suggests:<\/p>\n\n\n\n<p>Not latency optimization \u2014 but cost reduction.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">A\/B testing backlash<\/h3>\n\n\n\n<p>Different users received different product configurations silently.<\/p>\n\n\n\n<p>Trust took another hit.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">Migration is already happening<\/h2>\n\n\n\n<p>One interesting trend on Hacker News:<\/p>\n\n\n\n<p>People sharing their \u201cmigration stories.\u201d<\/p>\n\n\n\n<p>Some said they subconsciously switched to <a href=\"https:\/\/deepinsightai.io\/fr\/chatgpt-codex-vs-claude-code\/\" target=\"_blank\" rel=\"noreferrer noopener\">Codex<\/a> back in February \u2014 only now realizing Claude\u2019s decline may have pushed them.<\/p>\n\n\n\n<p>Others claim GPT-5.4 already outperformed Opus 4.6.<\/p>\n\n\n\n<p>Some are supplementing with MiniMax:<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>$40 \u2192 4500 messages per 5-hour cycle<\/li>\n\n\n\n<li>Full reasoning visibility<\/li>\n<\/ul>\n\n\n\n<h2 class=\"wp-block-heading\">The landscape has changed<\/h2>\n\n\n\n<p>Six months ago, \u201cuse Claude for coding\u201d was almost consensus.<\/p>\n\n\n\n<p>Now:<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>Codex reportedly has 4 million active users<\/li>\n\n\n\n<li>GPT-5.5 focuses on coding + computer operation<\/li>\n\n\n\n<li>OpenAI staff even describe it as a \u201cchief of staff\u201d model<\/li>\n<\/ul>\n\n\n\n<p>Claude didn\u2019t necessarily get worse.<\/p>\n\n\n\n<p>Others got better \u2014 and Claude stumbled at the worst possible moment.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">The window is shrinking<\/h2>\n\n\n\n<p>Anthropic now has less time than before to:<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>Fix bugs<\/li>\n\n\n\n<li>Rebuild trust<\/li>\n<\/ul>\n\n\n\n<p>Meanwhile:<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>GPT-5.5 is already out<\/li>\n\n\n\n<li><a href=\"https:\/\/deepinsightai.io\/fr\/deepseek-v4\/\" target=\"_blank\" rel=\"noreferrer noopener\">DeepSeek V4<\/a> is ready<\/li>\n<\/ul>\n\n\n\n<p>And people are waiting:<\/p>\n\n\n\n<p>Gemini, your move.<\/p>\n\n\n\n<p><\/p>","protected":false},"excerpt":{"rendered":"<p>Right around the release of GPT-5.5, Claude finally admitted it: Yes, the model got \u201cdumber,\u201d and all usage quotas have been reset. After over a month of denial, the intelligence drop bug finally came straight from Anthropic itself: Stack those three bugs together, and Claude\u2019s overall experience took a serious hit. Good thing competitors applied [&hellip;]<\/p>\n","protected":false},"author":1,"featured_media":2636,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"_seopress_robots_primary_cat":"none","_seopress_titles_title":"%%post_title%%","_seopress_titles_desc":"Anthropic confirms Claude\u2019s performance drop caused by 3 major bugs\u2014reasoning downgrade, cache failure, and prompt limits. Full breakdown, timeline, and user backlash after GPT-5.5 release.","_seopress_robots_index":"","_uag_custom_page_level_css":"","site-sidebar-layout":"default","site-content-layout":"","ast-site-content-layout":"default","site-content-style":"default","site-sidebar-style":"default","ast-global-header-display":"","ast-banner-title-visibility":"","ast-main-header-display":"","ast-hfb-above-header-display":"","ast-hfb-below-header-display":"","ast-hfb-mobile-header-display":"","site-post-title":"","ast-breadcrumbs-content":"","ast-featured-img":"","footer-sml-layout":"","ast-disable-related-posts":"","theme-transparent-header-meta":"","adv-header-id-meta":"","stick-header-meta":"","header-above-stick-meta":"","header-main-stick-meta":"","header-below-stick-meta":"","astra-migrate-meta-layouts":"set","ast-page-background-enabled":"default","ast-page-background-meta":{"desktop":{"background-color":"","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"tablet":{"background-color":"","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"mobile":{"background-color":"","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""}},"ast-content-background-meta":{"desktop":{"background-color":"var(--ast-global-color-5)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"tablet":{"background-color":"var(--ast-global-color-5)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"mobile":{"background-color":"var(--ast-global-color-5)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""}},"footnotes":""},"categories":[2,10],"tags":[],"class_list":["post-2630","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-ai-news","category-llm"],"uagb_featured_image_src":{"full":["https:\/\/deepinsightai.io\/wp-content\/uploads\/2026\/04\/Claude-Admits-Getting-Dumber-3-Critical-Bugs-Exposed-After-GPT-5.5-Launch.webp",2528,1696,false],"thumbnail":["https:\/\/deepinsightai.io\/wp-content\/uploads\/2026\/04\/Claude-Admits-Getting-Dumber-3-Critical-Bugs-Exposed-After-GPT-5.5-Launch-150x150.webp",150,150,true],"medium":["https:\/\/deepinsightai.io\/wp-content\/uploads\/2026\/04\/Claude-Admits-Getting-Dumber-3-Critical-Bugs-Exposed-After-GPT-5.5-Launch-300x201.webp",300,201,true],"medium_large":["https:\/\/deepinsightai.io\/wp-content\/uploads\/2026\/04\/Claude-Admits-Getting-Dumber-3-Critical-Bugs-Exposed-After-GPT-5.5-Launch-768x515.webp",768,515,true],"large":["https:\/\/deepinsightai.io\/wp-content\/uploads\/2026\/04\/Claude-Admits-Getting-Dumber-3-Critical-Bugs-Exposed-After-GPT-5.5-Launch-1024x687.webp",1024,687,true],"1536x1536":["https:\/\/deepinsightai.io\/wp-content\/uploads\/2026\/04\/Claude-Admits-Getting-Dumber-3-Critical-Bugs-Exposed-After-GPT-5.5-Launch-1536x1030.webp",1536,1030,true],"2048x2048":["https:\/\/deepinsightai.io\/wp-content\/uploads\/2026\/04\/Claude-Admits-Getting-Dumber-3-Critical-Bugs-Exposed-After-GPT-5.5-Launch-2048x1374.webp",2048,1374,true],"trp-custom-language-flag":["https:\/\/deepinsightai.io\/wp-content\/uploads\/2026\/04\/Claude-Admits-Getting-Dumber-3-Critical-Bugs-Exposed-After-GPT-5.5-Launch-18x12.webp",18,12,true]},"uagb_author_info":{"display_name":"Claude Carter","author_link":"https:\/\/deepinsightai.io\/fr\/author\/cloud-han03gmail-com\/"},"uagb_comment_info":0,"uagb_excerpt":"Right around the release of GPT-5.5, Claude finally admitted it: Yes, the model got \u201cdumber,\u201d and all usage quotas have been reset. After over a month of denial, the intelligence drop bug finally came straight from Anthropic itself: Stack those three bugs together, and Claude\u2019s overall experience took a serious hit. Good thing competitors applied\u2026","_links":{"self":[{"href":"https:\/\/deepinsightai.io\/fr\/wp-json\/wp\/v2\/posts\/2630","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/deepinsightai.io\/fr\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/deepinsightai.io\/fr\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/deepinsightai.io\/fr\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/deepinsightai.io\/fr\/wp-json\/wp\/v2\/comments?post=2630"}],"version-history":[{"count":3,"href":"https:\/\/deepinsightai.io\/fr\/wp-json\/wp\/v2\/posts\/2630\/revisions"}],"predecessor-version":[{"id":2650,"href":"https:\/\/deepinsightai.io\/fr\/wp-json\/wp\/v2\/posts\/2630\/revisions\/2650"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/deepinsightai.io\/fr\/wp-json\/wp\/v2\/media\/2636"}],"wp:attachment":[{"href":"https:\/\/deepinsightai.io\/fr\/wp-json\/wp\/v2\/media?parent=2630"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/deepinsightai.io\/fr\/wp-json\/wp\/v2\/categories?post=2630"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/deepinsightai.io\/fr\/wp-json\/wp\/v2\/tags?post=2630"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}