{"id":6335,"date":"2026-10-22T07:17:02","date_gmt":"2026-10-22T07:17:02","guid":{"rendered":"https:\/\/i10xnews.kinsta.cloud\/?p=6335"},"modified":"2026-10-22T07:17:02","modified_gmt":"2026-10-22T07:17:02","slug":"frontiercs-2-ai-agents-beat-published-research-code","status":"publish","type":"post","link":"https:\/\/i10x.ai\/news\/frontiercs-2-ai-agents-beat-published-research-code","title":{"rendered":"Only One AI Agent Beats Researchers&#8217; Own Code on Most FrontierCS 2 Tasks"},"content":{"rendered":"\n<h2 class=\"wp-block-heading\">\u26a1 Quick Take<\/h2>\n\n\n<ul class=\"wp-block-list\"><li><strong>The finding:<\/strong> FrontierCS 2 asks coding agents to rebuild a paper&#8217;s deleted contribution so that it beats the authors&#8217; own code. Across 48 tasks in 13 computer-science domains, Claude Opus 5.5 in Claude Code passed 58% of runs. Every other agent passed under half, and no agent passed any of 16 tasks.<\/li><li><strong>Why the bar is hard:<\/strong> A run passes only when the final patch beats the human reference on every metric and every held-out workload. Most runs that score higher than the reference on a combined score still miss at least one threshold.<\/li><li><strong>What it means:<\/strong> Agents can contribute to frontier CS R&amp;D, but the median task is about as hard as what the strongest agent can handle. Which agent leads also depends on the cost budget per run.<\/li><\/ul>\n\n\n<p class=\"wp-block-paragraph\">Most open-ended coding benchmarks hand an agent a problem statement and grade one program after the run. Real research looks different: you inherit a working repository, change one component, test on the paper&#8217;s workloads and are judged against the authors&#8217; own implementation. FrontierCS 2, a large multi-institution effort, builds that setting into a benchmark of 48 tasks and asks whether today&#8217;s LLM agents can clear it.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">\ud83d\udd2c What the study shows<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">Each task is built by <strong>subtraction<\/strong>. The authors take a paper&#8217;s released repository, delete the code that implements the paper&#8217;s contribution and keep that deleted code as a private human reference. The agent may edit only the files that held the contribution, submit patches for five hours and receive development-suite scores. The evaluator, the held-out final suite and the reference stay with the judge. At the end the agent names one patch as final, and that patch alone is graded on held-out workloads the agent never saw.<\/p>\n\n\n\n<figure class=\"wp-block-table\"><table><thead><tr><th>Agent (harness)<\/th><th>Pass rate<\/th><\/tr><\/thead><tbody><tr><td><strong>Claude Opus 5.5 (Claude Code)<\/strong><\/td><td><strong>58%<\/strong><\/td><\/tr><tr><td>GPT-6 Astra (Codex CLI)<\/td><td>44%<\/td><\/tr><tr><td>Gemini 4 Argon (Pi)<\/td><td>40%<\/td><\/tr><tr><td>GPT-6.1 Sol (Codex CLI)<\/td><td>35%<\/td><\/tr><tr><td>GLM 5.3 (ZCode)<\/td><td>25%<\/td><\/tr><tr><td>Kimi K3 (Kimi Code)<\/td><td>21%<\/td><\/tr><tr><td>Muse Wonder 1.0 (Muse Code)<\/td><td>21%<\/td><\/tr><tr><td>DeepSeek V4.1 Flash<\/td><td>15%<\/td><\/tr><tr><td>Muse Spark 1.3 \/ Qwen 3.8 Max \/ GLM 5.3 Flash<\/td><td>12%<\/td><\/tr><tr><td>Step 5 Preview (Step Code)<\/td><td>8%<\/td><\/tr><\/tbody><\/table><figcaption class=\"wp-element-caption\">Figure 2 of the paper: share of each agent&#8217;s 48 runs whose final submission beat the human reference on every metric and held-out workload (one run per task).<\/figcaption><\/figure>\n\n\n<ul class=\"wp-block-list\"><li><strong>Pass \u2260 higher score:<\/strong> Most runs whose combined final score exceeds the human reference still fall short on some metric or workload. Conversely, some passing runs score at or below the reference&#8217;s final score, because a pass needs every metric to hold at once.<\/li><li><strong>Sixteen tasks stay unsolved:<\/strong> No agent passed any of 16 tasks. On SVG-EAR, for example, GPT-6 Astra was faster than the reference and closer to full attention on average, yet still failed image-quality bands on two held-out clips.<\/li><li><strong>Shared capability scale:<\/strong> The FrontierCS Capabilities Index (FCI) places agents and tasks on one scale. Claude Opus 5.5 leads FCI; the median task sits at about the strongest agent&#8217;s FCI.<\/li><li><strong>Budget changes the leader:<\/strong> Under a small cost cap per run, DeepSeek V4.1 Flash leads (with a caveat that its token costs are lower bounds). Claude Opus 5.5 leads at the largest budgets.<\/li><\/ul>\n\n\n<figure class=\"wp-block-image size-large\"><img loading=\"lazy\" decoding=\"async\" width=\"1344\" height=\"768\" src=\"https:\/\/i10xnews.kinsta.cloud\/wp-content\/uploads\/2026\/10\/chart-pass-rate.webp\" alt=\"Horizontal bar chart of FrontierCS 2 pass rates: Claude Opus 5.5 at 58 percent, GPT-6 Astra 44, Gemini 4 Argon 40, GPT-6.1 Sol 35, then eight other agents from 25 down to 8 percent.\" class=\"wp-image-6334\" srcset=\"https:\/\/i10x.ai\/news\/wp-content\/uploads\/2026\/10\/chart-pass-rate.webp 1344w, https:\/\/i10x.ai\/news\/wp-content\/uploads\/2026\/10\/chart-pass-rate-300x171.webp 300w, https:\/\/i10x.ai\/news\/wp-content\/uploads\/2026\/10\/chart-pass-rate-1024x585.webp 1024w, https:\/\/i10x.ai\/news\/wp-content\/uploads\/2026\/10\/chart-pass-rate-768x439.webp 768w\" sizes=\"auto, (max-width: 1344px) 100vw, 1344px\" \/><figcaption class=\"wp-element-caption\">Pass rates across 48 tasks. i10X chart based on data from the FrontierCS Team (2026), arXiv:2610.29932, Figure 2.<\/figcaption><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\">Three case studies show what a pass captures and what it misses. On QuarterPIR (private information retrieval), Muse Wonder 1.0 replaced heavy reply payloads with two RAID-6-style checksums and shrank lookup size far more than other passing runs, yet a pass counts the same for a small or large saving. On BBQ (FPGA packet scheduler), Claude Opus 5.5 and Gemini 4 Argon both beat the reference on throughput, but Claude&#8217;s circuit used more lookup tables; the final score rewarded speed and ignored size. On SVG-EAR, speed was easy; staying inside every image-quality band was not.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">\u2705 Why it matters: graded against the authors&#8217; own code<\/h2>\n\n\n<ul class=\"wp-block-list\"><li><strong>Oracle is the paper&#8217;s code:<\/strong> Unlike SWE-bench tests or Kaggle leaderboards, the bar is the authors&#8217; released implementation measured on the paper&#8217;s own workloads.<\/li><li><strong>Held-out final suite:<\/strong> Development scores guide the agent but do not count. Only the agent-chosen final patch is graded on held-out seeds, queries or graphs.<\/li><li><strong>13 domains, not only ML:<\/strong> Tasks cover databases, HPC, cryptography, networks, architecture, security, robotics, programming languages and more, spanning 48 public tasks plus a private set of more than 100.<\/li><li><strong>Isolated sandboxes:<\/strong> The agent has no network except a driver that forwards model calls and patches, so it cannot download the authors&#8217; code. GPU tasks keep agent and verifier GPUs separate.<\/li><\/ul>\n\n\n<h2 class=\"wp-block-heading\">\ud83e\uddea Method in brief<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>Agents:<\/strong> 12 model\u2013harness pairs, each run once per task (48 runs each). Harnesses include Claude Code, Codex CLI, Pi, ZCode, Muse Code, DeepSeek Harness, Kimi Code, Qwen Code and Step Code. <strong>Time:<\/strong> five hours per task. <strong>Pass:<\/strong> beat the human reference on every metric across every held-out workload. <strong>FCI:<\/strong> an item-response model fitted to per-metric outcomes and score margins over the reference (624 runs including an anchor agent), with 90% bootstrap intervals. Two Gemini 4 Argon runs refused by the provider count as failures.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">\u2753 What remains open<\/h2>\n\n\n<ul class=\"wp-block-list\"><li><strong>One run per cell:<\/strong> With a single attempt per agent\u2013task pair, pass rates have no per-cell confidence intervals; FCI intervals help, but individual task difficulty estimates are uncertain.<\/li><li><strong>Side costs ignored:<\/strong> Pass and final score can miss circuit size, bytes saved or other trade-offs that matter in practice. The authors say benchmarks like this will need to report those next to the pass.<\/li><li><strong>Harness is part of the agent:<\/strong> Results attribute performance to a model\u2013harness pair. Separating model strength from harness design would need a factorial design.<\/li><li><strong>Private holdout:<\/strong> More than 100 tasks remain private for future evaluation, so today&#8217;s leaderboard can overfit the public 48.<\/li><\/ul>\n\n\n<h2 class=\"wp-block-heading\">\ud83e\udded Context: from SWE-bench to research repositories<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">FrontierCS 2 sits between SWE-bench-style environment grading and ML-only research agents such as PaperBench, ResearchGym and MLS-Bench. The closest design is MLS-Bench, which also asks agents to improve a named component; FrontierCS 2 extends that idea beyond machine learning and sets the bar with the authors&#8217; own code. It also connects to earlier i10X coverage of scientific agents that <a href=\"\/news\/frontierchallenge-scientific-workflow-completion\">finish only a fraction of end-to-end workflows<\/a>.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">\ud83d\udee0\ufe0f What it means for teams evaluating coding agents<\/h2>\n\n\n<ul class=\"wp-block-list\"><li><strong>Measure against real baselines:<\/strong> Beating a weak starter or a test suite is easier than beating the paper&#8217;s own code. Vendor claims should say which bar they cleared.<\/li><li><strong>Budget matters:<\/strong> The cheapest agent is not the same as the best agent at unlimited spend. Pick a cost cap before ranking.<\/li><li><strong>Inspect side costs:<\/strong> A pass that blows up circuit size or ignores memory can be the wrong win for production systems.<\/li><\/ul>\n\n\n<p class=\"wp-block-paragraph\">For models in the leaderboard, see our <a href=\"\/news\/company\/anthropic\">Anthropic<\/a>, <a href=\"\/news\/company\/open-ai\">OpenAI<\/a>, <a href=\"\/news\/company\/google\">Google<\/a> and <a href=\"\/news\/company\/deepseek\">DeepSeek<\/a> hubs, or more <a href=\"\/news\/category\/benchmarks-evals\">benchmarks and evals<\/a>.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">\u270d\ufe0f About the analysis<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">This article is based on the full text of the preprint &#8220;FrontierCS 2: A Benchmark for Computer Science R&amp;D&#8221; (arXiv:2610.29932v1, submitted October 21, 2026), including appendices on task sources and the FCI model. All numbers come from the paper; i10X did not rerun the experiments. The paper uses arXiv&#8217;s non-exclusive license, so charts were redrawn by i10X from Figure 2. Code is at github.com\/FrontierCS\/FrontierCS-2.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">\ud83d\udd2d i10x Perspective<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">FrontierCS 2 is a useful correction to leaderboard hype. Agents can rebuild real research contributions, but only the top systems clear a majority of tasks, and many &#8220;better scores&#8221; still fail a pass. The interesting next step is not another 48 tasks, but reporting the side costs that labs actually care about \u2014 size, latency, memory \u2014 next to the pass. Until then, treat a FrontierCS 2 pass as evidence of research-level coding under a hard bar, not as a general claim that agents can replace research engineers.<\/p>\n\n","protected":false},"excerpt":{"rendered":"<p>FrontierCS 2 deletes a paper&#8217;s contribution from its repository and asks agents to beat the authors&#8217; own code. Claude Opus 5.5 passes 58% of 48 tasks; every other agent passes under half, and 16 tasks stay unsolved.<\/p>\n","protected":false},"author":2,"featured_media":6333,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"paper_title":"FrontierCS 2: A Benchmark for Computer Science R&D","paper_url":"https:\/\/arxiv.org\/abs\/2610.29932","pdf_url":"https:\/\/arxiv.org\/pdf\/2610.29932v1","doi":"","arxiv_id":"2610.29932","authors":"Qiuyang Mang, Wenhao Chai, Runyuan He, Dongruixuan Li, Zihan Zheng, Yujun He, and the FrontierCS Team (61 authors)","institutions":"UC Berkeley, Princeton University, University of Waterloo, NYU, Georgia Tech, Stanford, MIT, Scale AI and collaborating universities","paper_date":"2026-10-21","venue":"arXiv (cs.LG), version 1","peer_reviewed":false,"code_url":"https:\/\/github.com\/FrontierCS\/FrontierCS-2","data_url":"https:\/\/github.com\/FrontierCS\/FrontierCS-2","funding_coi":"Thanks Modal, Laude Institute Slingshot, Scale AI compute credits and Kaggle. No conflicts of interest stated.","figure_license":"Paper figures not reused; chart redrawn by i10X from Figure 2","paper_license":"arXiv non-exclusive distribution license","replication":"partial","study_models":"Claude Opus 5.5, GPT-6 Astra, GPT-6.1 Sol, Gemini 4 Argon, GLM 5.3 (+ Flash), Kimi K3, Muse Wonder 1.0, Muse Spark 1.3, DeepSeek V4.1 Flash, Qwen 3.8 Max, Step 5 Preview","study_sample":"48 public tasks across 13 CS domains; 12 agents \u00d7 1 run per task; private holdout of >100 tasks","footnotes":"","_yoast_wpseo_primary_category":""},"categories":[53,52],"tags":[],"companies":[39,46,37,38],"products":[51,49],"class_list":["post-6335","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-benchmarks-evals","category-studies","company-anthropic","company-deepseek","company-google","company-open-ai","product-claude","product-gemini"],"yoast_head":"<!-- This site is optimized with the Yoast SEO plugin v28.7 - https:\/\/yoast.com\/product\/yoast-seo-wordpress\/ -->\n<title>FrontierCS 2: Only Claude Opus 5.5 Clears Most Tasks - i10X News<\/title>\n<meta name=\"description\" content=\"New benchmark: agents must beat researchers&#039; own published code. Claude Opus 5.5 passes 58% of 48 CS R&amp;D tasks; 16 tasks stay unsolved by every agent.\" \/>\n<meta name=\"robots\" content=\"index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1\" \/>\n<link rel=\"canonical\" href=\"https:\/\/i10x.ai\/news\/frontiercs-2-ai-agents-beat-published-research-code\" \/>\n<meta property=\"og:locale\" content=\"en_US\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"FrontierCS 2: Only Claude Opus 5.5 Clears Most Tasks - i10X News\" \/>\n<meta property=\"og:description\" content=\"New benchmark: agents must beat researchers&#039; own published code. Claude Opus 5.5 passes 58% of 48 CS R&amp;D tasks; 16 tasks stay unsolved by every agent.\" \/>\n<meta property=\"og:url\" content=\"https:\/\/i10x.ai\/news\/frontiercs-2-ai-agents-beat-published-research-code\" \/>\n<meta property=\"og:site_name\" content=\"i10X News\" \/>\n<meta property=\"article:published_time\" content=\"2026-10-22T07:17:02+00:00\" \/>\n<meta property=\"og:image\" content=\"https:\/\/i10x.ai\/news\/wp-content\/uploads\/2026\/10\/frontiercs-2-ai-agents-beat-published-research-code-title.webp\" \/>\n\t<meta property=\"og:image:width\" content=\"1344\" \/>\n\t<meta property=\"og:image:height\" content=\"768\" \/>\n\t<meta property=\"og:image:type\" content=\"image\/webp\" \/>\n<meta name=\"author\" content=\"Christopher Ort\" \/>\n<meta name=\"twitter:card\" content=\"summary_large_image\" \/>\n<meta name=\"twitter:creator\" content=\"@i10x_ai\" \/>\n<meta name=\"twitter:site\" content=\"@i10x_ai\" \/>\n<meta name=\"twitter:label1\" content=\"Written by\" \/>\n\t<meta name=\"twitter:data1\" content=\"Christopher Ort\" \/>\n\t<meta name=\"twitter:label2\" content=\"Est. reading time\" \/>\n\t<meta name=\"twitter:data2\" content=\"6 minutes\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\\\/\\\/schema.org\",\"@graph\":[{\"@type\":\"NewsArticle\",\"@id\":\"https:\\\/\\\/i10x.ai\\\/news\\\/frontiercs-2-ai-agents-beat-published-research-code#article\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/i10x.ai\\\/news\\\/frontiercs-2-ai-agents-beat-published-research-code\"},\"author\":{\"name\":\"Christopher Ort\",\"@id\":\"https:\\\/\\\/news.i10x.ai\\\/#\\\/schema\\\/person\\\/d410bb7ee72febefca2c9c369939219a\"},\"headline\":\"Only One AI Agent Beats Researchers&#8217; Own Code on Most FrontierCS 2 Tasks\",\"datePublished\":\"2026-10-22T07:17:02+00:00\",\"mainEntityOfPage\":{\"@id\":\"https:\\\/\\\/i10x.ai\\\/news\\\/frontiercs-2-ai-agents-beat-published-research-code\"},\"wordCount\":1260,\"commentCount\":0,\"publisher\":{\"@id\":\"https:\\\/\\\/news.i10x.ai\\\/#organization\"},\"image\":{\"@id\":\"https:\\\/\\\/i10x.ai\\\/news\\\/frontiercs-2-ai-agents-beat-published-research-code#primaryimage\"},\"thumbnailUrl\":\"https:\\\/\\\/i10x.ai\\\/news\\\/wp-content\\\/uploads\\\/2026\\\/10\\\/frontiercs-2-ai-agents-beat-published-research-code-title.webp\",\"articleSection\":[\"Benchmarks &amp; Evals\",\"Studies\"],\"inLanguage\":\"en-US\",\"potentialAction\":[{\"@type\":\"CommentAction\",\"name\":\"Comment\",\"target\":[\"https:\\\/\\\/i10x.ai\\\/news\\\/frontiercs-2-ai-agents-beat-published-research-code#respond\"]}],\"dateModified\":\"2026-10-22T07:17:02+00:00\"},{\"@type\":\"WebPage\",\"@id\":\"https:\\\/\\\/i10x.ai\\\/news\\\/frontiercs-2-ai-agents-beat-published-research-code\",\"url\":\"https:\\\/\\\/i10x.ai\\\/news\\\/frontiercs-2-ai-agents-beat-published-research-code\",\"name\":\"FrontierCS 2: Only Claude Opus 5.5 Clears Most Tasks - i10X News\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/news.i10x.ai\\\/#website\"},\"primaryImageOfPage\":{\"@id\":\"https:\\\/\\\/i10x.ai\\\/news\\\/frontiercs-2-ai-agents-beat-published-research-code#primaryimage\"},\"image\":{\"@id\":\"https:\\\/\\\/i10x.ai\\\/news\\\/frontiercs-2-ai-agents-beat-published-research-code#primaryimage\"},\"thumbnailUrl\":\"https:\\\/\\\/i10x.ai\\\/news\\\/wp-content\\\/uploads\\\/2026\\\/10\\\/frontiercs-2-ai-agents-beat-published-research-code-title.webp\",\"datePublished\":\"2026-10-22T07:17:02+00:00\",\"description\":\"New benchmark: agents must beat researchers' own published code. Claude Opus 5.5 passes 58% of 48 CS R&D tasks; 16 tasks stay unsolved by every agent.\",\"breadcrumb\":{\"@id\":\"https:\\\/\\\/i10x.ai\\\/news\\\/frontiercs-2-ai-agents-beat-published-research-code#breadcrumb\"},\"inLanguage\":\"en-US\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\\\/\\\/i10x.ai\\\/news\\\/frontiercs-2-ai-agents-beat-published-research-code\"]}]},{\"@type\":\"ImageObject\",\"inLanguage\":\"en-US\",\"@id\":\"https:\\\/\\\/i10x.ai\\\/news\\\/frontiercs-2-ai-agents-beat-published-research-code#primaryimage\",\"url\":\"https:\\\/\\\/i10x.ai\\\/news\\\/wp-content\\\/uploads\\\/2026\\\/10\\\/frontiercs-2-ai-agents-beat-published-research-code-title.webp\",\"contentUrl\":\"https:\\\/\\\/i10x.ai\\\/news\\\/wp-content\\\/uploads\\\/2026\\\/10\\\/frontiercs-2-ai-agents-beat-published-research-code-title.webp\",\"width\":1344,\"height\":768,\"caption\":\"Illustration: i10X\"},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\\\/\\\/i10x.ai\\\/news\\\/frontiercs-2-ai-agents-beat-published-research-code#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"Home\",\"item\":\"https:\\\/\\\/news.i10x.ai\\\/\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"Only One AI Agent Beats Researchers&#8217; Own Code on Most FrontierCS 2 Tasks\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\\\/\\\/news.i10x.ai\\\/#website\",\"url\":\"https:\\\/\\\/news.i10x.ai\\\/\",\"name\":\"i10X News\",\"description\":\"Stay updated with the latest news and announcements from i10X.\",\"publisher\":{\"@id\":\"https:\\\/\\\/news.i10x.ai\\\/#organization\"},\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\\\/\\\/news.i10x.ai\\\/?s={search_term_string}\"},\"query-input\":{\"@type\":\"PropertyValueSpecification\",\"valueRequired\":true,\"valueName\":\"search_term_string\"}}],\"inLanguage\":\"en-US\"},{\"@type\":\"Organization\",\"@id\":\"https:\\\/\\\/news.i10x.ai\\\/#organization\",\"name\":\"i10X\",\"url\":\"https:\\\/\\\/news.i10x.ai\\\/\",\"logo\":{\"@type\":\"ImageObject\",\"inLanguage\":\"en-US\",\"@id\":\"https:\\\/\\\/news.i10x.ai\\\/#\\\/schema\\\/logo\\\/image\\\/\",\"url\":\"https:\\\/\\\/i10xnews.kinsta.cloud\\\/wp-content\\\/uploads\\\/2026\\\/10\\\/i10x-logo-180.png\",\"contentUrl\":\"https:\\\/\\\/i10xnews.kinsta.cloud\\\/wp-content\\\/uploads\\\/2026\\\/10\\\/i10x-logo-180.png\",\"width\":180,\"height\":180,\"caption\":\"i10X\"},\"image\":{\"@id\":\"https:\\\/\\\/news.i10x.ai\\\/#\\\/schema\\\/logo\\\/image\\\/\"},\"sameAs\":[\"https:\\\/\\\/x.com\\\/i10x_ai\",\"https:\\\/\\\/www.linkedin.com\\\/company\\\/i10x-ai\",\"https:\\\/\\\/instagram.com\\\/i10x.ai\",\"https:\\\/\\\/www.youtube.com\\\/@i10X_ai_official\"]},{\"@type\":\"Person\",\"@id\":\"https:\\\/\\\/news.i10x.ai\\\/#\\\/schema\\\/person\\\/d410bb7ee72febefca2c9c369939219a\",\"name\":\"Christopher Ort\",\"image\":{\"@type\":\"ImageObject\",\"inLanguage\":\"en-US\",\"@id\":\"https:\\\/\\\/i10x.ai\\\/news\\\/wp-content\\\/uploads\\\/2026\\\/10\\\/christopher-ort-192.webp\",\"url\":\"https:\\\/\\\/i10x.ai\\\/news\\\/wp-content\\\/uploads\\\/2026\\\/10\\\/christopher-ort-192.webp\",\"contentUrl\":\"https:\\\/\\\/i10x.ai\\\/news\\\/wp-content\\\/uploads\\\/2026\\\/10\\\/christopher-ort-192.webp\",\"caption\":\"Christopher Ort\"},\"description\":\"Christopher Ort handles SEO, AI agent workflows and PR & partnerships at i10x.ai, the all-in-one AI workspace built by i10X Global Pte. Ltd. in Singapore. He is a graduate of Friedrich-Alexander-Universit\u00e4t Erlangen-N\u00fcrnberg (FAU). He is the author of \u201cThe Style Penalty\u201d, a study published on SSRN that analyses 1,576 evaluation decisions by four frontier AI models to show how AI-written CVs fare in automated hiring screens; the research has been covered by heise online and t3n.\",\"url\":\"https:\\\/\\\/i10x.ai\\\/news\\\/author\\\/christopher-ort\",\"sameAs\":[\"https:\\\/\\\/i10x.ai\\\/blog\\\/author\\\/christopher-ort\"],\"worksFor\":{\"@id\":\"https:\\\/\\\/i10x.ai\\\/news\\\/#organization\"},\"alumniOf\":{\"@type\":\"EducationalOrganization\",\"name\":\"Friedrich-Alexander-Universit\u00e4t Erlangen-N\u00fcrnberg (FAU)\",\"url\":\"https:\\\/\\\/www.fau.eu\"}}]}<\/script>\n<!-- \/ Yoast SEO plugin. -->","yoast_head_json":{"title":"FrontierCS 2: Only Claude Opus 5.5 Clears Most Tasks - i10X News","description":"New benchmark: agents must beat researchers' own published code. Claude Opus 5.5 passes 58% of 48 CS R&D tasks; 16 tasks stay unsolved by every agent.","robots":{"index":"index","follow":"follow","max-snippet":"max-snippet:-1","max-image-preview":"max-image-preview:large","max-video-preview":"max-video-preview:-1"},"canonical":"https:\/\/i10x.ai\/news\/frontiercs-2-ai-agents-beat-published-research-code","og_locale":"en_US","og_type":"article","og_title":"FrontierCS 2: Only Claude Opus 5.5 Clears Most Tasks - i10X News","og_description":"New benchmark: agents must beat researchers' own published code. Claude Opus 5.5 passes 58% of 48 CS R&D tasks; 16 tasks stay unsolved by every agent.","og_url":"https:\/\/i10x.ai\/news\/frontiercs-2-ai-agents-beat-published-research-code","og_site_name":"i10X News","article_published_time":"2026-10-22T07:17:02+00:00","og_image":[{"width":1344,"height":768,"url":"https:\/\/i10x.ai\/news\/wp-content\/uploads\/2026\/10\/frontiercs-2-ai-agents-beat-published-research-code-title.webp","type":"image\/webp"}],"author":"Christopher Ort","twitter_card":"summary_large_image","twitter_creator":"@i10x_ai","twitter_site":"@i10x_ai","twitter_misc":{"Written by":"Christopher Ort","Est. reading time":"6 minutes"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"NewsArticle","@id":"https:\/\/i10x.ai\/news\/frontiercs-2-ai-agents-beat-published-research-code#article","isPartOf":{"@id":"https:\/\/i10x.ai\/news\/frontiercs-2-ai-agents-beat-published-research-code"},"author":{"name":"Christopher Ort","@id":"https:\/\/news.i10x.ai\/#\/schema\/person\/d410bb7ee72febefca2c9c369939219a"},"headline":"Only One AI Agent Beats Researchers&#8217; Own Code on Most FrontierCS 2 Tasks","datePublished":"2026-10-22T07:17:02+00:00","mainEntityOfPage":{"@id":"https:\/\/i10x.ai\/news\/frontiercs-2-ai-agents-beat-published-research-code"},"wordCount":1260,"commentCount":0,"publisher":{"@id":"https:\/\/news.i10x.ai\/#organization"},"image":{"@id":"https:\/\/i10x.ai\/news\/frontiercs-2-ai-agents-beat-published-research-code#primaryimage"},"thumbnailUrl":"https:\/\/i10x.ai\/news\/wp-content\/uploads\/2026\/10\/frontiercs-2-ai-agents-beat-published-research-code-title.webp","articleSection":["Benchmarks &amp; Evals","Studies"],"inLanguage":"en-US","potentialAction":[{"@type":"CommentAction","name":"Comment","target":["https:\/\/i10x.ai\/news\/frontiercs-2-ai-agents-beat-published-research-code#respond"]}],"dateModified":"2026-10-22T07:17:02+00:00"},{"@type":"WebPage","@id":"https:\/\/i10x.ai\/news\/frontiercs-2-ai-agents-beat-published-research-code","url":"https:\/\/i10x.ai\/news\/frontiercs-2-ai-agents-beat-published-research-code","name":"FrontierCS 2: Only Claude Opus 5.5 Clears Most Tasks - i10X News","isPartOf":{"@id":"https:\/\/news.i10x.ai\/#website"},"primaryImageOfPage":{"@id":"https:\/\/i10x.ai\/news\/frontiercs-2-ai-agents-beat-published-research-code#primaryimage"},"image":{"@id":"https:\/\/i10x.ai\/news\/frontiercs-2-ai-agents-beat-published-research-code#primaryimage"},"thumbnailUrl":"https:\/\/i10x.ai\/news\/wp-content\/uploads\/2026\/10\/frontiercs-2-ai-agents-beat-published-research-code-title.webp","datePublished":"2026-10-22T07:17:02+00:00","description":"New benchmark: agents must beat researchers' own published code. Claude Opus 5.5 passes 58% of 48 CS R&D tasks; 16 tasks stay unsolved by every agent.","breadcrumb":{"@id":"https:\/\/i10x.ai\/news\/frontiercs-2-ai-agents-beat-published-research-code#breadcrumb"},"inLanguage":"en-US","potentialAction":[{"@type":"ReadAction","target":["https:\/\/i10x.ai\/news\/frontiercs-2-ai-agents-beat-published-research-code"]}]},{"@type":"ImageObject","inLanguage":"en-US","@id":"https:\/\/i10x.ai\/news\/frontiercs-2-ai-agents-beat-published-research-code#primaryimage","url":"https:\/\/i10x.ai\/news\/wp-content\/uploads\/2026\/10\/frontiercs-2-ai-agents-beat-published-research-code-title.webp","contentUrl":"https:\/\/i10x.ai\/news\/wp-content\/uploads\/2026\/10\/frontiercs-2-ai-agents-beat-published-research-code-title.webp","width":1344,"height":768,"caption":"Illustration: i10X"},{"@type":"BreadcrumbList","@id":"https:\/\/i10x.ai\/news\/frontiercs-2-ai-agents-beat-published-research-code#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https:\/\/news.i10x.ai\/"},{"@type":"ListItem","position":2,"name":"Only One AI Agent Beats Researchers&#8217; Own Code on Most FrontierCS 2 Tasks"}]},{"@type":"WebSite","@id":"https:\/\/news.i10x.ai\/#website","url":"https:\/\/news.i10x.ai\/","name":"i10X News","description":"Stay updated with the latest news and announcements from i10X.","publisher":{"@id":"https:\/\/news.i10x.ai\/#organization"},"potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/news.i10x.ai\/?s={search_term_string}"},"query-input":{"@type":"PropertyValueSpecification","valueRequired":true,"valueName":"search_term_string"}}],"inLanguage":"en-US"},{"@type":"Organization","@id":"https:\/\/news.i10x.ai\/#organization","name":"i10X","url":"https:\/\/news.i10x.ai\/","logo":{"@type":"ImageObject","inLanguage":"en-US","@id":"https:\/\/news.i10x.ai\/#\/schema\/logo\/image\/","url":"https:\/\/i10xnews.kinsta.cloud\/wp-content\/uploads\/2026\/10\/i10x-logo-180.png","contentUrl":"https:\/\/i10xnews.kinsta.cloud\/wp-content\/uploads\/2026\/10\/i10x-logo-180.png","width":180,"height":180,"caption":"i10X"},"image":{"@id":"https:\/\/news.i10x.ai\/#\/schema\/logo\/image\/"},"sameAs":["https:\/\/x.com\/i10x_ai","https:\/\/www.linkedin.com\/company\/i10x-ai","https:\/\/instagram.com\/i10x.ai","https:\/\/www.youtube.com\/@i10X_ai_official"]},{"@type":"Person","@id":"https:\/\/news.i10x.ai\/#\/schema\/person\/d410bb7ee72febefca2c9c369939219a","name":"Christopher Ort","image":{"@type":"ImageObject","inLanguage":"en-US","@id":"https:\/\/i10x.ai\/news\/wp-content\/uploads\/2026\/10\/christopher-ort-192.webp","url":"https:\/\/i10x.ai\/news\/wp-content\/uploads\/2026\/10\/christopher-ort-192.webp","contentUrl":"https:\/\/i10x.ai\/news\/wp-content\/uploads\/2026\/10\/christopher-ort-192.webp","caption":"Christopher Ort"},"description":"Christopher Ort handles SEO, AI agent workflows and PR & partnerships at i10x.ai, the all-in-one AI workspace built by i10X Global Pte. Ltd. in Singapore. He is a graduate of Friedrich-Alexander-Universit\u00e4t Erlangen-N\u00fcrnberg (FAU). He is the author of \u201cThe Style Penalty\u201d, a study published on SSRN that analyses 1,576 evaluation decisions by four frontier AI models to show how AI-written CVs fare in automated hiring screens; the research has been covered by heise online and t3n.","url":"https:\/\/i10x.ai\/news\/author\/christopher-ort","sameAs":["https:\/\/i10x.ai\/blog\/author\/christopher-ort"],"worksFor":{"@id":"https:\/\/i10x.ai\/news\/#organization"},"alumniOf":{"@type":"EducationalOrganization","name":"Friedrich-Alexander-Universit\u00e4t Erlangen-N\u00fcrnberg (FAU)","url":"https:\/\/www.fau.eu"}}]}},"_links":{"self":[{"href":"https:\/\/i10x.ai\/news\/wp-json\/wp\/v2\/posts\/6335","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/i10x.ai\/news\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/i10x.ai\/news\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/i10x.ai\/news\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/i10x.ai\/news\/wp-json\/wp\/v2\/comments?post=6335"}],"version-history":[{"count":1,"href":"https:\/\/i10x.ai\/news\/wp-json\/wp\/v2\/posts\/6335\/revisions"}],"predecessor-version":[{"id":6338,"href":"https:\/\/i10x.ai\/news\/wp-json\/wp\/v2\/posts\/6335\/revisions\/6338"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/i10x.ai\/news\/wp-json\/wp\/v2\/media\/6333"}],"wp:attachment":[{"href":"https:\/\/i10x.ai\/news\/wp-json\/wp\/v2\/media?parent=6335"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/i10x.ai\/news\/wp-json\/wp\/v2\/categories?post=6335"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/i10x.ai\/news\/wp-json\/wp\/v2\/tags?post=6335"},{"taxonomy":"company","embeddable":true,"href":"https:\/\/i10x.ai\/news\/wp-json\/wp\/v2\/companies?post=6335"},{"taxonomy":"product","embeddable":true,"href":"https:\/\/i10x.ai\/news\/wp-json\/wp\/v2\/products?post=6335"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}