--- title: Kimi K3 Beat Claude at Coding. It Also Hallucinates 51%... description: Kimi K3, a 2.8 trillion parameter open-weight model, outperformed Claude and GPT-5 in blind frontend coding benchmarks. However, independent evaluations... canonical: https://daily.dev/posts/kimi-k3-beat-claude-at-coding-it-also-hallucinates-51-of-the-time--dqhvdtiiu twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: Kimi K3 Beat Claude at Coding. It Also Hallucinates 51% of the Time. | daily.dev og:description: Kimi K3, a 2.8 trillion parameter open-weight model, outperformed Claude and GPT-5 in blind frontend coding benchmarks. However, independent evaluations... og:url: https://daily.dev/posts/kimi-k3-beat-claude-at-coding-it-also-hallucinates-51-of-the-time--dqhvdtiiu og:image: https://api.daily.dev/og/posts/DqhVdtiIu.png og:image:alt: Kimi K3 Beat Claude at Coding. It Also Hallucinates 51% of the Time. 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. # Kimi K3 Beat Claude at Coding. It Also Hallucinates 51% of the Time. **[Noor Mohammad](https://daily.dev/sources/ew9hy5zthevyyefagop4g)** · [@noormohammad](https://daily.dev/noormohammad) · 1 min read · 0 upvotes · 1 comments ## Summary Kimi K3, a 2.8 trillion parameter open-weight model, outperformed Claude and GPT-5 in blind frontend coding benchmarks. However, independent evaluations revealed a 51% hallucination rate — worse than its predecessor. The post fact-checks major claims about the model against primary sources, clarifies what 'open weights' actually means in practice, and offers guidance for developers deciding whether to build with it. ## Content Kimi K3: 2.8T params, largest open-weight model ever released, and in blind dev testing it beat Claude Fable 5 and GPT-5.6 Sol at frontend coding. Then independent evals landed the other number: 51% hallucination rate — worse than its own predecessor. Champion coder, confident liar, same model, same week. Hype threads gave you half; skeptics gave you the other half. This fact-checks every major claim against primary sources, covers why "open weights" ≠ what most people assume, and what a builder should actually do about it. 🔗 [https://medium.com/@reactjsbd/the-biggest-open-ai-model-ever-just-beat-claude-at-coding-it-also-hallucinates-half-the-time-83b50de39023?sharedUserId=reactjsbd](https://medium.com/@reactjsbd/the-biggest-open-ai-model-ever-just-beat-claude-at-coding-it-also-hallucinates-half-the-time-83b50de39023?sharedUserId=reactjsbd) ## Community discussion Top comments from developers on daily.dev. **@noormohammad** · 0 upvotes > The "open weights" caveat is the part most coverage skips — weights available ≠ open source, and at 2.8T params, who's actually self-hosting this anyway? For anyone who's tried K3: is the frontend coding win real in practice, or benchmark-shaped? And how are you handling the hallucination rate — verification layer, task gating, or avoiding it entirely? ## Similar posts on daily.dev - [I Tried Kimi K3 for a Week. Here’s What Happened.](https://daily.dev/posts/i-tried-kimi-k3-for-a-week-here-s-what-happened--ufveuen98) · Medium · 2 upvotes · 0 comments --- Tags: [#open-source](https://daily.dev/tags/open-source), [#claude](https://daily.dev/tags/claude), [#kimi](https://daily.dev/tags/kimi) [View this post on daily.dev](https://daily.dev/posts/kimi-k3-beat-claude-at-coding-it-also-hallucinates-51-of-the-time--dqhvdtiiu) ```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":"DiscussionForumPosting","mainEntityOfPage":"https://daily.dev/posts/kimi-k3-beat-claude-at-coding-it-also-hallucinates-51-of-the-time--dqhvdtiiu","headline":"Kimi K3 Beat Claude at Coding. It Also Hallucinates 51% of the Time.","text":"Kimi K3, a 2.8 trillion parameter open-weight model, outperformed Claude and GPT-5 in blind frontend coding benchmarks. However, independent evaluations revealed a 51% hallucination rate — worse than its predecessor. The post fact-checks major claims about the model against primary sources, clarifies what 'open weights' actually means in practice, and offers guidance for developers deciding whether to build with it.","url":"https://daily.dev/posts/kimi-k3-beat-claude-at-coding-it-also-hallucinates-51-of-the-time--dqhvdtiiu","datePublished":"2026-07-23T18:01:24.672Z","dateModified":"2026-09-13T20:59:29.091Z","author":{"@type":"Person","name":"Noor Mohammad","url":"https://daily.dev/noormohammad","image":"https://media.daily.dev/image/upload/s--3w4n0n-6--/f_auto/v1754054115/avatars/avatar_eW9hy5zTHEVyYeFaGOP4G?_a=BAMClqZW0","description":"Software Engineer | React, Next.js & scalable web apps | Writing on AI, blockchain & emerging tech","interactionStatistic":{"@type":"InteractionCounter","interactionType":{"@type":"EndorseAction"},"userInteractionCount":170}},"image":"https://media.daily.dev/image/upload/s--8OBBIE6p--/f_auto/v1784829684/posts/DqhVdtiIu?_a=BAMAMicg0","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":0},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":1}],"comment":[{"@type":"Comment","text":"The “open weights” caveat is the part most coverage skips — weights available ≠ open source, and at 2.8T params, who’s actually self-hosting this anyway? For anyone who’s tried K3: is the frontend coding win real in practice, or benchmark-shaped? And how are you handling the hallucination rate — verification layer, task gating, or avoiding it entirely?","datePublished":"2026-07-23T18:01:40.464Z","url":"https://daily.dev/posts/DqhVdtiIu#c-nIokbUWKA","author":{"@type":"Person","name":"Noor Mohammad","url":"https://daily.dev/noormohammad","image":"https://media.daily.dev/image/upload/s--3w4n0n-6--/f_auto/v1754054115/avatars/avatar_eW9hy5zTHEVyYeFaGOP4G?_a=BAMClqZW0"}}],"isPartOf":{"@type":"WebPage","url":"https://daily.dev/sources/ew9hy5zthevyyefagop4g","name":"Noor Mohammad"}} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Noor Mohammad","item":"https://daily.dev/sources/ew9hy5zthevyyefagop4g"},{"@type":"ListItem","position":3,"name":"Kimi K3 Beat Claude at Coding. It Also Hallucinates 51% of the Time."}]} ```