Claude’s new model is more ‘honest’ when it messes up | The VergeSkip to main contentThe homepageThe VergeThe Verge logo.The VergeThe Verge logo.TechReviewsScienceEntertainmentAIPolicyNotificationsNotificationsHamburger Navigation ButtonThe homepageThe VergeThe Verge logo.NotificationsNotificationsHamburger Navigation ButtonNavigation DrawerThe VergeThe Verge logo.Login / Sign UpcloseCloseSearchTechExpandAmazonAppleFacebookGoogleMicrosoftSamsungBusinessSee all techReviewsExpandSmart Home ReviewsPhone ReviewsTablet ReviewsHeadphone ReviewsSee all reviewsScienceExpandSpaceEnergyEnvironmentHealthSee all scienceEntertainmentExpandTV ShowsMoviesAudioSee all entertainmentAIExpandOpenAIAnthropicSee all AIPolicyExpandAntitrustPoliticsLawSecuritySee all policyGadgetsExpandLaptopsPhonesTVsHeadphonesSpeakersWearablesSee all gadgetsVerge ShoppingExpandBuying GuidesDealsGift GuidesSee all shoppingGamingExpandXboxPlayStationNintendoSee all gamingStreamingExpandDisneyHBONetflixYouTubeCreatorsSee all streamingTransportationExpandElectric CarsAutonomous CarsRide-sharingScootersSee all transportationFeaturesVerge VideoExpandTikTokYouTubeInstagramPodcastsExpandDecoderThe VergecastVersion HistoryNewslettersArchivesStoreVerge Product UpdatesSubscribeFacebookThreadsInstagramYoutubeRSSThe VergeThe Verge logo.Claude’s new model is more ‘honest’ when it messes upNotificationsNotificationsComments DrawerNotificationsCommentsLoading commentsGetting the conversation ready...AICloseAIPosts from this topic will be added to your daily email digest and your homepage feed.FollowFollowSee All AINewsCloseNewsPosts from this topic will be added to your daily email digest and your homepage feed.FollowFollowSee All NewsAnthropicCloseAnthropicPosts from this topic will be added to your daily email digest and your homepage feed.FollowFollowSee All AnthropicClaude’s new model is more ‘honest’ when it messes upOpus 4.8 is launching on Thursday.Opus 4.8 is launching on Thursday.by Jay PetersCloseJay PetersSenior ReporterPosts from this author will be added to your daily email digest and your homepage feed.FollowFollowSee All by Jay PetersMay 28, 2026, 5:00 PM UTCLinkShareGiftImage: Cath Virginia / The Verge, Getty ImagesJay PetersCloseJay PetersPosts from this author will be added to your daily email digest and your homepage feed.FollowFollowSee All by Jay Peters is a senior reporter covering technology, gaming, and more. He joined The Verge in 2019 after nearly two years at Techmeme.Anthropic is releasing Claude Opus 4.8 on Thursday, and the company is touting the model’s “honesty.”According to Anthropic, it trains “all [its] models to be honest — for instance, to avoid making claims that they can’t support.” But it notes that “a general problem with AI models is that they sometimes jump to conclusions, confidently presenting their work as making progress despite thin evidence.”The AI lab claims that early testers have found that Opus 4.8 “is more likely to flag uncertainties about its work and less likely to make unsupported claims.” In the company’s evaluations, Opus 4.8 is “around 4x less likely than its predecessor to allow flaws in code it’s written to pass unremarked.”In addition to the honesty improvements, with Opus 4.8, users can direct the amount of effort Claude puts into a task. Higher-effort responses will use more tokens, giving users the option of lower-effort responses if they don’t want to burn through their rate limits as quickly.Anthropic is also launching a feature called “dynamic workflows” in research preview, which the company says will let Claude “take on even bigger tasks.” With dynamic workflows, “Claude can plan the work and then run hundreds of parallel subagents in a single session (and with Opus 4.8, the agents can run for even longer). It then verifies its outputs before reporting back to the user.”Follow topics and authors from this story to see more like this in your personalized homepage feed and to receive email updates.Jay PetersCloseJay PetersSenior ReporterPosts from this author will be added to your daily email digest and your homepage feed.FollowFollowSee All by Jay PetersAICloseAIPosts from this topic will be added to your daily email digest and your homepage feed.FollowFollowSee All AIAnthropicCloseAnthropicPosts from this topic will be added to your daily email digest and your homepage feed.FollowFollowSee All AnthropicNewsCloseNewsPosts from this topic will be added to your daily email digest and your homepage feed.FollowFollowSee All NewsMost PopularMost PopularValve raises Steam Deck prices by more than $200The golden age of handheld gaming is already overSony’s first RGB TV is a statement pieceKia’s flagship EV has a battery problemSony is offering up to 50 percent off some of our favorite PS5 gamesThe Verge DailyA free daily digest of the news that matters most.Email (required)Sign UpBy submitting your email, you agree to our Terms and Privacy Notice. This site is protected by reCAPTCHA and the Google Privacy Policy and Terms of Service apply.Advertiser Content FromThis is the title for the native adMore in AIThese new iOS 27 renders hint at Siri’s big redesignCNN sues Perplexity over ‘verbatim’ copycat articlesPlayRivian’s software chief thinks you don’t need CarPlay or buttonsYouTube will let you ask AI to make a custom video feedAI tried to bury this politician — now people have actually heard of himRobinhood will let your AI agent trade stocks and make (or lose) lots of moneyThese new iOS 27 renders hint at Siri’s big redesignStevie BonifieldTwo hours agoCNN sues Perplexity over ‘verbatim’ copycat articlesEmma RothTwo hours agoPlayRivian’s software chief thinks you don’t need CarPlay or buttonsNilay Patel2:00 PM UTCYouTube will let you ask AI to make a custom video feedJess Weatherbed8:49 AM UTCAI tried to bury this politician — now people have actually heard of himTina NguyenMay 27Robinhood will let your AI agent trade stocks and make (or lose) lots of moneyEmma RothMay 27Advertiser Content FromThis is the title for the native adTop Stories6:30 AM UTCThe golden age of handheld gaming is already over12:00 PM UTCKia’s flagship EV has a battery problemMay 27Roids were all the rage at the Enhanced Games12:00 PM UTCThey’ve finally made the Oura Ring smaller and lighterMay 27The AI fight brewing inside The New York Times17 minutes agoHere’s where you can preorder the new Oura Ring 5The VergeThe Verge logo.FacebookThreadsInstagramYoutubeRSSContactTip UsCommunity GuidelinesArchivesAboutEthics StatementHow We Rate and Review ProductsCookie SettingsTerms of UsePrivacy NoticeCookie PolicyLicensing FAQAccessibilityPlatform Status© 2026 Vox Media, LLC. All Rights ReservedNotifications DrawerThe VergeThe Verge logo.Sign in to see your notifications or create an account to join the conversation.Sign in

Anthropic is launching its new model, Claude Opus 4.8, emphasizing the system's increased "honesty" as a core feature. The company's training methodology aims to instill honesty across its models by instructing them to avoid making claims they cannot support. This addresses a recognized general issue with artificial intelligence models, which often jump to conclusions and present their output as progress despite lacking sufficient evidence. Early evaluations of Opus 4.8 suggest that this goal has been partially achieved; testers found that the model is more likely to flag uncertainties regarding its output and is less inclined to make unsupported assertions. Furthermore, the improvements in reliability are quantified by noting that Opus 4.8 is approximately four times less likely than its predecessor to permit errors in the code it generates to go undetected. Beyond enhancing factual veracity, the new iteration introduces user control over the computational effort expended by Claude on a given task. Users now have the option to specify higher-effort responses, which utilize more tokens, or opt for lower-effort responses to manage token usage and rate limits more efficiently. To enable handling more complex requests, Anthropic is also rolling out a feature called dynamic workflows in research preview. This feature allows Claude to manage complex tasks by first planning the overall work and then executing hundreds of parallel subagents within a single session, with the agents independently verifying their outputs before reporting the final result to the user.