{"id":38017,"date":"2025-06-21T13:10:13","date_gmt":"2025-06-21T05:10:13","guid":{"rendered":"https:\/\/www.1ai.net\/?p=38017"},"modified":"2025-06-21T13:10:13","modified_gmt":"2025-06-21T05:10:13","slug":"kimi-researcher-%e6%b7%b1%e5%ba%a6%e7%a0%94%e7%a9%b6%e6%a8%a1%e5%9e%8b%e5%bc%80%e5%90%af%e5%86%85%e6%b5%8b%ef%bc%9a%e5%8f%af%e7%94%9f%e6%88%90%e6%98%93%e8%bf%bd%e6%ba%af%e7%9a%84%e4%b8%87%e5%ad%97","status":"publish","type":"post","link":"https:\/\/www.1ai.net\/en\/38017.html","title":{"rendered":"Kimi-Researcher deep research model opens for internal testing: generates easily traceable 10,000-word reports"},"content":{"rendered":"<p>June 21, 2011 - 1AI has learned from the<a href=\"https:\/\/www.1ai.net\/en\/tag\/%e6%9c%88%e4%b9%8b%e6%9a%97%e9%9d%a2\" title=\"[Sees articles with labels]\" target=\"_blank\" >Dark Side of the Moon<\/a> <a href=\"https:\/\/www.1ai.net\/en\/tag\/kimi\" title=\"[View articles tagged with [Kimi]]\" target=\"_blank\" >Kimi<\/a> Publicis has learned that Kimi's first <a href=\"https:\/\/www.1ai.net\/en\/tag\/agent\" title=\"[View articles tagged with [Agent]]\" target=\"_blank\" >Agent<\/a>\uff08<a href=\"https:\/\/www.1ai.net\/en\/tag\/%e6%99%ba%e8%83%bd%e4%bd%93\" title=\"[View articles tagged with [intelligent body]]\" target=\"_blank\" >Agent<\/a>\uff09<a href=\"https:\/\/www.1ai.net\/en\/tag\/kimi-researcher\" title=\"_Other Organiser\" target=\"_blank\" >Kimi-Researcher<\/a> Opens on the 20th<strong>local<a href=\"https:\/\/www.1ai.net\/en\/tag\/%e7%81%b0%e5%ba%a6%e6%b5%8b%e8%af%95\" title=\"[Sees articles with [grey test] labels]\" target=\"_blank\" >Grayscale test<\/a><\/strong>.<\/p>\n<p><img loading=\"lazy\" decoding=\"async\" class=\"alignnone size-full wp-image-38018\" title=\"035c536aj00sy6wzg004dd000u000csp\" src=\"https:\/\/www.1ai.net\/wp-content\/uploads\/2025\/06\/035c536aj00sy6wzg004dd000u000csp.jpg\" alt=\"035c536aj00sy6wzg004dd000u000csp\" width=\"1080\" height=\"460\" \/><\/p>\n<p>Kimi-Researcher is a new generation of agent model trained based on end-to-end agentic RL technology, and is also an agent product specially designed for deep research tasks. Later, Dark Side of the Moon will gradually open source Kimi-Researcher's basic pre-trained models and reinforcement learning models.<\/p>\n<p>For each problem, Kimi-Researcher autonomously plans the task execution process and ultimately delivers complete results:<\/p>\n<ul>\n<li><strong>Clarification<\/strong>: Proactive rhetorical questioning in understanding the problem and constructing a clearer problem space;<\/li>\n<li><strong>reflect in depth<\/strong>: An average of 23 steps of reasoning per task to autonomously sort out and address needs;<\/li>\n<li><strong>Proactive search<\/strong>For each task, an average of 74 keywords were planned, 206 URLs were found, and the top 3.2% contents with the highest information quality were judged and filtered by the model to eliminate redundant and low-quality information;<\/li>\n<li><strong>Invoke the tools and deliver the results<\/strong>: Autonomously invoke the browser, code and other tools to process raw data, automatically generate analysis conclusions, and complete the delivery end-to-end.<\/li>\n<\/ul>\n<p>To ensure the quality of the output and the coverage of the information, Kimi-Researcher uses the<strong>asynchronous execution<\/strong>way, spending more time progressively reasoning, retrieving and writing content.<\/p>\n<p><img loading=\"lazy\" decoding=\"async\" class=\"alignnone size-full wp-image-38020\" title=\"f7e7a38dj00sy6wz900c1d000u000f0m\" src=\"https:\/\/www.1ai.net\/wp-content\/uploads\/2025\/06\/f7e7a38dj00sy6wz900c1d000u000f0m.jpg\" alt=\"f7e7a38dj00sy6wz900c1d000u000f0m\" width=\"1080\" height=\"540\" \/><\/p>\n<p>Users will eventually receive 2 deliverables.<\/p>\n<p><strong>An informative and traceable in-depth research report<\/strong><\/p>\n<ul>\n<li>The average length of a report is between<strong>More than 10,000 words<\/strong>;<\/li>\n<li>An average of about 26 high-quality, traceable sources are cited;<\/li>\n<li>All citations are embedded in the body of the text, clickable to jump and highlighting the original text for easy verification and traceability.<\/li>\n<\/ul>\n<p><img loading=\"lazy\" decoding=\"async\" class=\"alignnone size-full wp-image-38019\" title=\"1e58fcc0j00sy6wz9009fd000jc00d8m\" src=\"https:\/\/www.1ai.net\/wp-content\/uploads\/2025\/06\/1e58fcc0j00sy6wz9009fd000jc00d8m.jpg\" alt=\"1e58fcc0j00sy6wz9009fd000jc00d8m\" width=\"696\" height=\"476\" \/><\/p>\n<p><strong>An interactive, shareable and dynamic visualization report<\/strong><\/p>\n<ul>\n<li>Structured layout and mind maps make trends, exceptions and other important information visible at a glance;<\/li>\n<li>The overall structure and core conclusions can be quickly grasped without having to read the entire text;<\/li>\n<li>Support online link generation and sharing for easy display.<\/li>\n<\/ul>\n<p>Officially, Kimi-Researcher is in full compliance with the \"Humanity's Last Exam, HLE\" in the highly difficult benchmark designed for AI<strong>Zero structure, no flow design<\/strong>The scores for the settings are as follows:<\/p>\n<ul>\n<li>Pass@1 Accuracy: 26.9%<\/li>\n<li>Pass@4 Accuracy: 40.17%<\/li>\n<\/ul>\n<p>This performance exceeds Claude 4 Opus (10.71 TP3T), Gemini 2.5 Pro (21.61 TP3T), is slightly higher than OpenAI Deep Research (26.61 TP3T), and is tied with Gemini-Pro's Deep Research Agent (26.91 TP3T), which This is one of the highest levels known to date. In the xbench benchmark released by Sequoia China, an AI capability evaluation system aligned with real task scenarios, Kimi-Researcher achieved an average pass rate of 69% in the DeepSearch task, ahead of other models in the list.<\/p>","protected":false},"excerpt":{"rendered":"<p>On June 21, 1AI learned from Kimi Public, the dark side of the month, that Kimi's first Agent (intelligence) Kimi-Researcher had started small-scale greyscale tests on June 20. Kimi-Researcher is a new generation of Agent models based on end-to-end autonomous RL technical training and is also an Agent product for in-depth research missions. The dark side of the moon will then gradually open the Kimi-Researcher Basic Pre-Training Model and the Enhanced Learning Model. For every problem, Kimi-Researcher will plan the mission<\/p>","protected":false},"author":1,"featured_media":0,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[146],"tags":[1405,1814,7006,1355,1168,705],"collection":[],"class_list":["post-38017","post","type-post","status-publish","format-standard","hentry","category-news","tag-agent","tag-kimi","tag-kimi-researcher","tag-1355","tag-1168","tag-705"],"acf":[],"_links":{"self":[{"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/posts\/38017","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/comments?post=38017"}],"version-history":[{"count":0,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/posts\/38017\/revisions"}],"wp:attachment":[{"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/media?parent=38017"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/categories?post=38017"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/tags?post=38017"},{"taxonomy":"collection","embeddable":true,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/collection?post=38017"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}