{"id":41592,"date":"2025-08-22T11:32:06","date_gmt":"2025-08-22T03:32:06","guid":{"rendered":"https:\/\/www.1ai.net\/?p=41592"},"modified":"2025-08-22T11:32:06","modified_gmt":"2025-08-22T03:32:06","slug":"deepseek-v3-1-%e6%ad%a3%e5%bc%8f%e5%8f%91%e5%b8%83%ef%bc%8c%e5%ae%98%e6%96%b9%e8%af%a6%e8%a7%a3%e8%bf%88%e5%90%91-ai-agent-%e6%97%b6%e4%bb%a3%e7%9a%84%e7%ac%ac%e4%b8%80%e6%ad%a5","status":"publish","type":"post","link":"https:\/\/www.1ai.net\/en\/41592.html","title":{"rendered":"DeepSeek-V3.1 Released, Officials Explain First Steps Toward AI Agent Era"},"content":{"rendered":"<p>August 22, 2011 - Depth Seeker officials yesterday officially released the<strong><a href=\"https:\/\/www.1ai.net\/en\/tag\/deepseek\" title=\"[View articles tagged with [DeepSeek]]\" target=\"_blank\" >DeepSeek<\/a>-V3.1<\/strong>.<\/p>\n<p><img loading=\"lazy\" decoding=\"async\" class=\"alignnone size-full wp-image-41593\" title=\"7a6943bfj00t1dlri00yjd000sg00izp\" src=\"https:\/\/www.1ai.net\/wp-content\/uploads\/2025\/08\/7a6943bfj00t1dlri00yjd000sg00izp.jpg\" alt=\"7a6943bfj00t1dlri00yjd000sg00izp\" width=\"1024\" height=\"683\" \/><\/p>\n<p>This upgrade contains the following major changes:<\/p>\n<ul>\n<li><strong>Hybrid Reasoning Architecture<\/strong>: A model supports both thinking and non-thinking modes;<\/li>\n<li><strong>More efficient thinking<\/strong>: DeepSeek-V3.1-Think gives answers in less time than DeepSeek-R1-0528;<\/li>\n<li><strong>stronger <a href=\"https:\/\/www.1ai.net\/en\/tag\/agent\" title=\"[View articles tagged with [Agent]]\" target=\"_blank\" >Agent<\/a> abilities<\/strong>: Through Post-Training optimization, the new model shows a large improvement in performance in the tool-use vs. smart-body task.<\/li>\n<\/ul>\n<p>Official App and Web Models<strong>Synchronized upgrade to DeepSeek-V3.1<\/strong>. Users can access this information through the<strong>\"Thinking Deeply.\"<\/strong>button to enable free switching between thinking mode and non-thinking mode.<\/p>\n<p>The DeepSeek API has also been upgraded, with deepseek-chat corresponding to the<strong>logical thinking<\/strong>The following is a list of the most common types of data that can be used by the Deepseek-reasoner.<strong>Thinking Patterns.<\/strong>The API Beta interface also supports strict mode Function Calling to ensure that the output Function satisfies the schema definition.<\/p>\n<p>In addition, Depth Seeking increases the number of<strong>Anthropic API<\/strong>format support, allowing users to plug the capabilities of DeepSeek-V3.1 into the<strong>Claude Code<\/strong>Framing.<\/p>\n<p><strong>Tool Call \/ Smartbody Support Enhancements<\/strong><\/p>\n<p>programmed intelligence<\/p>\n<p>DeepSeek-V3.1 shows significant improvement over the previous DeepSeek family of models in the Code Repair Evaluation SWE and the Complex Tasks in Command Line Terminal Environment (Terminal-Bench) test.<\/p>\n<p>Search for Intelligentsia<\/p>\n<p>DeepSeek-V3.1 has made significant improvements in several search evaluation metrics. DeepSeek-V3.1 has significantly outperformed R1-0528 on the complex search test (browsecomp) and the multidisciplinary expert-level problem test (HLE), which require multi-step reasoning.<\/p>\n<p>Thinking about efficiency gains<\/p>\n<p>The official test results of Deep Seeker show that after the thought chain compression training, the average performance of V3.1-Think is on par with that of R1-0528 for all tasks with a reduced number of output tokens 20%-50%.<\/p>\n<p>Meanwhile, the output length of V3.1 in non-thinking mode is also effectively controlled, and compared with DeepSeek-V3-0324, it can maintain the same model performance with significantly reduced output length.<\/p>\n<p><strong>API &amp; Model Open Source<\/strong><\/p>\n<p>modeling open source<\/p>\n<p>The Base model of V3.1 has been re-trained on the basis of V3, and a total of 840B tokens have been trained. both the Base model and the post-training model have been open-sourced at Huggingface and Magic Hitch.<\/p>\n<p>1AI Attached open source address:<\/p>\n<p><strong>Base model:<\/strong><\/p>\n<p>Hugging Face:<\/p>\n<ul>\n<li>https:\/\/huggingface.co\/deepseek-ai\/DeepSeek-V3.1-Base<\/li>\n<\/ul>\n<p>Magic Hitch:<\/p>\n<ul>\n<li>https:\/\/modelscope.cn\/models\/deepseek-ai\/DeepSeek-V3.1-Base<\/li>\n<\/ul>\n<p><strong>Post-training models:<\/strong><\/p>\n<p>Hugging Face:<\/p>\n<ul>\n<li>https:\/\/huggingface.co\/deepseek-ai\/DeepSeek-V3.1<\/li>\n<\/ul>\n<p>Magic Hitch:<\/p>\n<ul>\n<li>https:\/\/modelscope.cn\/models\/deepseek-ai\/DeepSeek-V3.1<\/li>\n<\/ul>\n<p>Note that DeepSeek-V3.1 uses the parameter accuracy of UE8M0 FP8 Scale. In addition, V3.1 has made significant adjustments to the splitter and chat template, which are significantly different from DeepSeek-V3. Users with deployment needs are recommended to read the new version of the documentation carefully.<\/p>\n<p><strong>price adjustment<\/strong><\/p>\n<p>Depth Seeking will be held at<strong>September 6, 2025 from midnight Beijing time<\/strong>The following adjustments are made to the prices of DeepSeek Open Platform API interface calls:<\/p>\n<ul>\n<li>Implementation of the new version of the price list;<\/li>\n<\/ul>\n<p>DeepSeek-V3.1 API<br \/>\nInput.<br \/>\n0.5 $\/million tokens (cache misses)<br \/>\n4$\/million tokens (cache hits)<br \/>\nOutput.<br \/>\n12 per million tokens<\/p>\n<ul>\n<li>Eliminate the night time hours discount.<\/li>\n<\/ul>\n<p>Until September 6, all API services will continue to be offered at the<strong>Original price policy billing<\/strong>In addition, users can continue to enjoy the current offer. At the same time, in order to better meet the user's calling needs, depth seeking has further expanded the API service resources.<\/p>","protected":false},"excerpt":{"rendered":"<p>August 22nd, DeepSeek officially released DeepSeek-V3.1 yesterday. The upgrade includes the following major changes: Hybrid reasoning architecture: one model supports both thinking and non-thinking modes; higher thinking efficiency: compared with DeepSeek-R1-0528, DeepSeek-V3.1-Think can give answers in a shorter time; stronger agent capability: through Post-Training optimization, the performance of the new model has been improved in tool usage and intelligent body tasks. Stronger Agent Capability: Through Post-Training optimization, the performance of the new model in tool usage and intelligent body tasks has been greatly improved. The official App and web models have been upgraded to DeepSeek-V3.1, and users can use the \"Deep Think\" button to realize thinking and non-thinking modes.<\/p>","protected":false},"author":1,"featured_media":0,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[146],"tags":[1405,3606],"collection":[],"class_list":["post-41592","post","type-post","status-publish","format-standard","hentry","category-news","tag-agent","tag-deepseek"],"acf":[],"_links":{"self":[{"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/posts\/41592","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=41592"}],"version-history":[{"count":0,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/posts\/41592\/revisions"}],"wp:attachment":[{"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/media?parent=41592"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/categories?post=41592"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/tags?post=41592"},{"taxonomy":"collection","embeddable":true,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/collection?post=41592"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}