{"id":27448,"date":"2025-01-21T10:51:40","date_gmt":"2025-01-21T02:51:40","guid":{"rendered":"https:\/\/www.1ai.net\/?p=27448"},"modified":"2025-01-21T10:51:40","modified_gmt":"2025-01-21T02:51:40","slug":"deepseek-r1-%e6%a8%a1%e5%9e%8b%e5%8f%91%e5%b8%83%ef%bc%8c%e6%80%a7%e8%83%bd%e5%af%b9%e6%a0%87-openai-o1-%e6%ad%a3%e5%bc%8f%e7%89%88","status":"publish","type":"post","link":"https:\/\/www.1ai.net\/en\/27448.html","title":{"rendered":"DeepSeek-R1 model released, performance benchmarked against OpenAI o1 release"},"content":{"rendered":"<p><a href=\"https:\/\/www.1ai.net\/en\/tag\/%e5%b9%bb%e6%96%b9%e9%87%8f%e5%8c%96\" title=\"[Sees articles with [Fantasy Quantification] labels]\" target=\"_blank\" >quantification by phantom<\/a>Deep Inquisition, an AI company under the umbrella of<a href=\"https:\/\/www.1ai.net\/en\/tag\/deepseek\" title=\"[View articles tagged with [DeepSeek]]\" target=\"_blank\" >DeepSeek<\/a>The DeepSeek-R1 model was officially released on January 20, and the model weights were open-sourced at the same time.<\/p>\n<p>According to the official introduction, DeepSeek-R1 uses reinforcement learning techniques on a large scale in the post-training phase, which greatly improves the model inference ability with only very little labeled data.<strong>Performance is comparable to the official OpenAI o1 version on tasks such as math, code, and natural language reasoning.<\/strong><\/p>\n<p>DeepSeek claims that<strong>DeepSeek-R1 distillation miniatures outperform OpenAI o1-mini<\/strong>DeepSeek, while open-sourcing two 660B models, DeepSeek-R1-Zero and DeepSeek-R1, has distilled the output of DeepSeek-R1 and open-sourced six smaller models to the community, of which the 32B and 70B models are benchmarked against the OpenAI o1-mini in a number of capabilities.<\/p>\n<p>Log in to the DeepSeek website or official app, open the \"Deep Thinking\" mode, and you can call the latest version of DeepSeek-R1 to complete all kinds of reasoning tasks.<\/p>\n<p><img loading=\"lazy\" decoding=\"async\" class=\"alignnone size-full wp-image-27449\" title=\"d94ad670j00sqf3wv0019d000tz00fum\" src=\"https:\/\/www.1ai.net\/wp-content\/uploads\/2025\/01\/d94ad670j00sqf3wv0019d000tz00fum.jpg\" alt=\"d94ad670j00sqf3wv0019d000tz00fum\" width=\"1079\" height=\"570\" \/><\/p>\n<p>The DeepSeek-R1 API service is priced at $1 (cache hits)\/$4 (cache misses) per million input tokens and $16 per million output tokens.<\/p>\n<p>1AI notes that DeepSeek has made all of its DeepSeek-R1 training technology publicly available, and while releasing and open-sourcing R1, it has simultaneously made the following adjustments at the protocol license level:<\/p>\n<ul>\n<li><strong>Model Open Source License Harmonized use of MIT<\/strong>We have introduced the DeepSeek License to provide licenses for the open source community. We have introduced DeepSeek License to provide authorization for the open source community in view of the characteristics of large model open source and reference to the current prevailing practice in the industry, but the practice shows that the non-standard open source License may instead increase the cost of understanding for the developers. For this reason, this time, our open source repository (including model weights) uniformly adopts the standardized and relaxed MIT License, completely open source, without restricting commercial use, no need to apply.<\/li>\n<li><strong>The product agreement specifies that \"model distillation\" is possible.<\/strong>We have decided to support users to perform \"model distillation\". In order to further promote open source and sharing of technology, we have decided to support \"model distillation\" by users. We have updated the user agreement of our online product to explicitly allow users to utilize model outputs and train other models through model distillation.<\/li>\n<\/ul>","protected":false},"excerpt":{"rendered":"<p>DeepSeek officially launched the DeepSeek-R1 model on January 20th under the Fantasy Quantification flag and synchronized the open source model weights. According to official accounts, DeepSeek-R1 used intensive learning technology on a large scale in the post-training phase, and in the case of very little data, model reasoning was greatly enhanced. On tasks such as mathematics, code, natural language reasoning, performance against shoulder OpenAI o1 official. DeepSeek states that DeepSeek-R1 distillation small models exceed OpenAI o1-mini. DeepSeek at open source DeepSeek-R1-Zero and DeepSeek-R1 two 6<\/p>","protected":false},"author":1,"featured_media":0,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[146],"tags":[3606,5600],"collection":[],"class_list":["post-27448","post","type-post","status-publish","format-standard","hentry","category-news","tag-deepseek","tag-5600"],"acf":[],"_links":{"self":[{"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/posts\/27448","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=27448"}],"version-history":[{"count":0,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/posts\/27448\/revisions"}],"wp:attachment":[{"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/media?parent=27448"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/categories?post=27448"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/tags?post=27448"},{"taxonomy":"collection","embeddable":true,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/collection?post=27448"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}