--- title: Gemini 4 Carbon takes #1, but it needs testing | daily.dev description: Google released Gemini 4 Carbon just two weeks after its previous model, Argon, claiming the #1 spot on the newly updated v5.0 Artificial Analysis Intelligence... canonical: https://daily.dev/posts/gemini-4-carbon-takes-1-but-it-needs-testing-yj9bfnnih twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: Gemini 4 Carbon takes #1, but it needs testing | daily.dev og:description: Google released Gemini 4 Carbon just two weeks after its previous model, Argon, claiming the #1 spot on the newly updated v5.0 Artificial Analysis Intelligence... og:url: https://daily.dev/posts/gemini-4-carbon-takes-1-but-it-needs-testing-yj9bfnnih og:image: https://api.daily.dev/og/posts/yJ9BfNNiH.png og:image:alt: Gemini 4 Carbon takes #1, but it needs testing 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. # Gemini 4 Carbon takes #1, but it needs testing **[Trends](https://daily.dev/sources/trends)** · 2 min read · 2 upvotes · 0 comments ## Summary Google released Gemini 4 Carbon just two weeks after its previous model, Argon, claiming the #1 spot on the newly updated v5.0 Artificial Analysis Intelligence Index with a score of 59. Carbon posts strong cybersecurity scores: 88.2% on SEC-bench Pro for finding vulnerabilities, 54.8% on ExploitGym for exploitation, and 71.0% on CWE-bench for patching. Google also reports gains in coding and agentic tasks, plus reduced misalignment and fabrication rates in internal testing. The model is rolling out to cyber defenders first rather than to everyone. All the headline cybersecurity and alignment numbers come from Google's own internal reporting, not independent testing. ## Content Google just put Gemini 4 Carbon at #1 on the Artificial Analysis Intelligence Index v5.0, and the replies are already splitting into "comeback season" and "read the fine print." The believers are loud. @YiTayML didn't bother with nuance: "Gemini 4 Carbon is the best frontier model right now!" @swyx went further, reopening @antigravity "for the first time in a long time" and declaring that "if google actually pulls this off it is officially comeback season with Meta, Grok and now GDM being cool again." He also called a field with roughly twice as many frontier labs as last year an "absolutely bizarre counter example to the bitter lesson." The skeptics are quieter but pointed. @scaling01 reads the timing as strategy: "Google strategically shipped Carbon before OpenAI and Anthropic drop their next gen monsters," and he thinks Google is "still ~2 months behind." A FrontierSWE result he shared backs the caution. Carbon scores 61.2%, beaten by GPT-6 Astra (65.5) and Opus 5.5 (62.3), among others. The arena numbers are mixed too, per @testingcatalog: - Text Arena: #1 at 1528 points - Agent Arena: #2, behind Opus 5.5, though it sets a new Pareto frontier - Web Dev Arena: #4, behind Opus 5.5, GPT-6 Astra and Sonnet 5.5 (still a 89 Elo jump) So "best model" depends on which board you read. Index and text chat, yes. Coding and agents, not quite. ## What Google is actually claiming The pitch is cybersecurity. Carbon scores 59 on the AA index and leads three security benchmarks: SEC-bench Pro for finding vulnerabilities (88.2%), ExploitGym for exploitation (54.8%) and CWE-bench for patching (71.0%). Google also says it resists 100% of evaluated prompt injection attacks and cuts unsupported claims about 2x versus Argon, which shipped just two weeks earlier. Pricing matches Argon at $2 per million input tokens and $10 per million output. Rollout starts with select cyber defenders, so most people can't poke at it yet. @testingcatalog's summary of the mood: "It needs testing!" That's the real gap in the conversation. The leaderboard victory lap is happening before most developers have touched the model, and the sharpest coding and agent benchmarks are the ones where Anthropic and OpenAI still hold the top spots. ## Questions this post answers ### What cybersecurity benchmark scores did Google report for Gemini 4 Carbon? Google reported 88.2% on SEC-bench Pro for finding vulnerabilities, 54.8% on ExploitGym for exploitation, and 71.0% on CWE-bench for patching. These are Google's own internal benchmark results, not independently verified figures. Google also claims the model ranks #1 on the newly updated v5.0 Artificial Analysis Intelligence Index with a score of 59. _Developers weighing frontier model claims can track independent verification of benchmarks like these on daily.dev._ ### Is Gemini 4 Carbon available to all users? No, Gemini 4 Carbon is rolling out first to cyber defenders rather than to the general public. This staged rollout reflects the dual-use nature of a model that scores highly on finding, exploiting, and patching vulnerabilities, since the same capability could be misused for attacks. _Teams deciding when to adopt a new frontier model can follow rollout details like this on daily.dev._ ### How long after Gemini 4 Argon did Google release Gemini 4 Carbon? Google released Gemini 4 Carbon just two weeks after Argon, an unusually fast iteration cadence for a frontier model. The rapid pace raises questions about whether the underlying gains are as substantial as the version jump suggests, or whether naming is doing some of the work. _Anyone tracking frontier model release cadence can follow updates like this on daily.dev._ ## Community take How the wider developer community reacted, aggregated from 2 discussions and 513 comments across hackernews, x (as of 2026-10-29). **TL;DR:** The community is split on whether Gemini 4 Carbon's #1 ranking is meaningful, since it's not yet publicly available, loses to Opus 5.5 on agent/coding benchmarks, and trails on Web Dev Arena, while some see fast progress and compelling pricing. Gemini 4 Carbon: The discussion is dominated by jokes and complaints about Google's confusing element-based naming scheme, with a secondary thread of skepticism about Google's habit of announcing models without immediate availability; actual capability discussion is mixed, with some finding Argon (the prior model) weak in coding benchmarks while others report positive experiences. **Sentiment:** 13% positive · 31% mixed · 56% skeptical **The case for** - Several note the rapid pace of improvement between Argon and Carbon (claimed +9 points on the Artificial Analysis index in weeks). - Some report Argon provides exceptional value at its price point, especially for software engineering work. - Some highlight Carbon's much lower cost per agent task versus Opus 5.5 as the more important metric than raw rank. - One commenter found Argon's feedback and optimization suggestions better than competing models in a specific test. - A few see the narrowing gap on the coding agent index (Carbon close behind Opus 5.5) as a meaningful signal for Google. **The pushback** - Many point out the model hasn't actually shipped or is not publicly accessible yet, calling the #1 claim premature. - The element-based naming scheme (Argon, Carbon, Bismuth, etc.) is widely mocked as confusing and non-intuitive compared to competitors' size-ordered naming. - Multiple commenters note it ranks #2 in Agent Arena and #4 in Web Dev Arena, arguing those benchmarks matter more for real coding work than Text Arena. - Many are frustrated that Google announces models long before they're actually available to the public. - Some argue leaderboard rankings are now so short-lived (weekly cycles) that being #1 means little. **By community** - hackernews (skeptical): Most engagement centers on mocking the naming convention and expressing frustration over delayed availability, with mixed real-world performance reports on the prior Argon model. - x (mixed): Reactions split between excitement over the #1 ranking and skepticism about whether it matters given the model isn't publicly available and trails on agent/web-dev benchmarks. **Hottest debate:** Whether a #1 ranking on Text Arena/intelligence index is meaningful when the model lags behind Opus 5.5 on agent and coding benchmarks and isn't yet publicly accessible. **Open questions** - How will Carbon perform in real multi-hour agent workflows and large repos versus benchmark scores? - How will Carbon actually perform in real-world coding agent battles versus benchmark claims? - Is the intelligence index gap driven by raw model improvements or test-time compute tricks? - Will Carbon be broadly available soon, or will it follow Google's pattern of delayed rollout after announcement? - When will the model actually be publicly accessible, and at what price? **Highlights** > @scaling01 they didn’t ship anything. > — [John\_lussier\_ on x · 47 points](https://x.com/John_lussier_/status/2115505635647475878) > @swyx @antigravity @saranormous the chart literally says no publicly available api. so everyone's crowning a model they can't call yet > — [elian\_mcc on x · 2 points](https://x.com/elian_mcc/status/2115624967203181012) > @swyx @antigravity @saranormous one thing worth poking in your review. AA has carbon's hallucination rate going from 4% on argon to 11%. still second lowest though, opus 5.5 and astra are both at 50%. and it uses 123k output tokens a task vs opus at 173k > — [aimanhasnoname on x · 1 points](https://x.com/aimanhasnoname/status/2115633308637774086) > @KazuRimuru @testingcatalog Carbon = Fable = Deep Think > — [AmirHAskari on x](https://x.com/AmirHAskari/status/2115509416770330997) > @swyx @antigravity @saranormous The coding-agent gap closing is the real signal — Carbon within ~3 points of Opus 5.5 on that agent index after years of Google trailing there. Curious what Antigravity feels like on a real multi-file refactor vs Claude Code; bench wins don't always survive tool-loop friction. > — [SIGNoffical on x](https://x.com/SIGNoffical/status/2115626406059417951) **Source threads** - [hackernews](https://news.ycombinator.com/item?id=50276792) · 391 points · 361 comments - [x](https://x.com/swyx/status/2115624308567658713) · 0 points · 64 comments - [x](https://x.com/scaling01/status/2115502223593746913) · 0 points · 63 comments - [x](https://x.com/testingcatalog/status/2115503561085718955) · 1 points · 25 comments - [x](https://x.com/scaling01/status/2115560979379769420) · 0 points · 0 comments ## Similar posts on daily.dev - [Gemini 4 Argon debuts with a unified 1M input/output token limit](https://daily.dev/posts/gemini-4-argon-debuts-with-a-unified-1m-input-output-token-limit-uqdxppv6j) · The New Stack · 9 upvotes · 2 comments --- Tags: [#ai](https://daily.dev/tags/ai), [#llm](https://daily.dev/tags/llm), [#google](https://daily.dev/tags/google), [#google-gemini](https://daily.dev/tags/google-gemini), [#ai-security](https://daily.dev/tags/ai-security) [View this post on daily.dev](https://daily.dev/posts/gemini-4-carbon-takes-1-but-it-needs-testing-yj9bfnnih) ```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":"Gemini 4 Carbon takes #1, but it needs testing","url":"https://daily.dev/posts/gemini-4-carbon-takes-1-but-it-needs-testing-yj9bfnnih","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/gemini-4-carbon-takes-1-but-it-needs-testing-yj9bfnnih"},"datePublished":"2026-10-28T17:44:53.569Z","dateModified":"2026-10-29T03:29:53.723Z","description":"Google released Gemini 4 Carbon just two weeks after its previous model, Argon, claiming the #1 spot on the newly updated v5.0 Artificial Analysis Intelligence...","image":"https://pbs.twimg.com/media/HVu_woRWoAAAhOy.jpg","thumbnailUrl":"https://pbs.twimg.com/media/HVu_woRWoAAAhOy.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/gemini-4-carbon-takes-1-but-it-needs-testing-yj9bfnnih","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":2},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"ai,llm,google,google-gemini,ai-security","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":"Gemini 4 Carbon takes #1, but it needs testing"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/gemini-4-carbon-takes-1-but-it-needs-testing-yj9bfnnih#faq","mainEntity":[{"@type":"Question","name":"What cybersecurity benchmark scores did Google report for Gemini 4 Carbon?","acceptedAnswer":{"@type":"Answer","text":"Google reported 88.2% on SEC-bench Pro for finding vulnerabilities, 54.8% on ExploitGym for exploitation, and 71.0% on CWE-bench for patching. These are Google's own internal benchmark results, not independently verified figures. Google also claims the model ranks #1 on the newly updated v5.0 Artificial Analysis Intelligence Index with a score of 59. Developers weighing frontier model claims can track independent verification of benchmarks like these on daily.dev."}},{"@type":"Question","name":"Is Gemini 4 Carbon available to all users?","acceptedAnswer":{"@type":"Answer","text":"No, Gemini 4 Carbon is rolling out first to cyber defenders rather than to the general public. This staged rollout reflects the dual-use nature of a model that scores highly on finding, exploiting, and patching vulnerabilities, since the same capability could be misused for attacks. Teams deciding when to adopt a new frontier model can follow rollout details like this on daily.dev."}},{"@type":"Question","name":"How long after Gemini 4 Argon did Google release Gemini 4 Carbon?","acceptedAnswer":{"@type":"Answer","text":"Google released Gemini 4 Carbon just two weeks after Argon, an unusually fast iteration cadence for a frontier model. The rapid pace raises questions about whether the underlying gains are as substantial as the version jump suggests, or whether naming is doing some of the work. Anyone tracking frontier model release cadence can follow updates like this on daily.dev."}}]} ```