{"id":1901308,"date":"2026-04-24T13:30:59","date_gmt":"2026-04-24T10:30:59","guid":{"rendered":"https:\/\/analyse.optim.biz\/?p=1901308"},"modified":"2026-04-24T13:30:59","modified_gmt":"2026-04-24T10:30:59","slug":"deepseek-previews-new-ai-model-that-closes-the-gap-with-frontier-models","status":"publish","type":"post","link":"https:\/\/analyse.optim.biz\/?p=1901308","title":{"rendered":"DeepSeek previews new AI model that \u2018closes the gap\u2019 with frontier models"},"content":{"rendered":"<p>[analyse_image type=&#8221;featured&#8221; src=&#8221;https:\/\/techcrunch.com\/wp-content\/uploads\/2026\/04\/GettyImages-2272184430.jpg?w=1024&#8243;]<\/p>\n<div class=\"entry-content wp-block-post-content is-layout-constrained wp-block-post-content-is-layout-constrained\">\n<p id=\"speakable-summary\" class=\"wp-block-paragraph\">Chinese AI lab DeepSeek has launched two preview versions of its newest large language model, DeepSeek V4, a much-awaited update to last year\u2019s V3.2 model and the accompanying R1 reasoning model that took the AI world by storm.<\/p>\n<p class=\"wp-block-paragraph\">The company says both DeepSeek V4 Flash and V4 Pro are mixture-of-experts models with context windows of 1 million tokens each \u2014 enough to allow large codebases or documents to be used in prompts. The mixture-of-experts approach involves activating only a certain number of parameters per task to lower inference costs.<\/p>\n<p class=\"wp-block-paragraph\">The Pro model has a total of 1.6 trillion parameters (49 billion active), which makes it the biggest open-weight model available, outstripping Moonshot AI\u2019s Kimi K 2.6 (1.1 trillion), MiniMax\u2019s M1 (456 billion), and more than double DeepSeek V3.2 (671 billion). The smaller, V4 Flash has 284 billion parameters (13 billion active).<\/p>\n<p class=\"wp-block-paragraph\">DeepSeek says both models are more efficient and performant than DeepSeek V3.2 due to architectural improvements, and have almost \u201cclosed the gap\u201d with current leading models, both open and closed, on reasoning benchmarks. <\/p>\n<p class=\"wp-block-paragraph\">The company claims its new V4-Pro-Max model outperforms its opensource peers across reasoning benchmarks, and outstrips OpenAI\u2019s GPT-5.2 and Gemini 3.0 Pro on some tasks. In coding competition benchmarks, DeepSeek said both V4 models\u2019 performance is \u201ccomparable to GPT-5.4.\u201d<\/p>\n<figure class=\"wp-block-image aligncenter size-large\"><img decoding=\"async\" loading=\"lazy\" height=\"469\" width=\"680\" src=\"https:\/\/techcrunch.com\/wp-content\/uploads\/2026\/04\/260424_18h44m20s_screenshot.png?w=680\" alt class=\"wp-image-3116128\"><figcaption class=\"wp-element-caption\"><span class=\"wp-block-image__credits\"><strong>Image Credits:<\/strong>DeepSeek<\/span><\/figcaption><\/figure>\n<p class=\"wp-block-paragraph\">However, the models seem to fall slightly behind frontier models in knowledge tests, specifically OpenAI\u2019s GPT-5.4 and Google\u2019s latest Gemini 3.1 Pro. This lag suggests a \u201cdevelopmental trajectory that trails state-of-the-art frontier models by approximately 3 to 6 months,\u201d the lab wrote.<\/p>\n<p class=\"wp-block-paragraph\">Both V4 Flash and V4 Pro support text only, unlike many of its closed-source peers, which offer support for understanding and generating audio, video, and images.<\/p>\n<div class=\"wp-block-techcrunch-inline-cta\">\n<div class=\"inline-cta__wrapper\">\n<div class=\"inline-cta__flag\">Techcrunch event<\/div>\n<div class=\"inline-cta__content\">\n<div class=\"inline-cta__header-container\">\n<div class=\"inline-cta__header-container-desktop\">\n<h3 class=\"inline-cta__header has-h-5-font-size\">Meet your next investor or portfolio startup at Disrupt<\/h3>\n<h4 class=\"inline-cta__subheader\">Your next round. Your next hire. Your next breakout opportunity. Find it at TechCrunch Disrupt 2026, where 10,000+ founders, investors, and tech leaders gather for three days of 250+ tactical sessions, powerful introductions, and market-defining innovation. Register now to save up to $410.<\/h4>\n<\/div>\n<div class=\"inline-cta__header-container-mobile\">\n<h3 class=\"inline-cta__header has-h-5-font-size\">Meet your next investor or portfolio startup at Disrupt<\/h3>\n<h4 class=\"inline-cta__subheader\">Your next round. Your next hire. Your next breakout opportunity. Find it at TechCrunch Disrupt 2026, where 10,000+ founders, investors, and tech leaders gather for three days of 250+ tactical sessions, powerful introductions, and market-defining innovation. Register now to save up to $410.<\/h4>\n<\/div>\n<\/div>\n<div class=\"inline-cta__event-info\"><span class=\"inline-cta__location\">San Francisco, CA<\/span><span class=\"inline-cta__separator\">|<\/span><span class=\"inline-cta__date\">October 13-15, 2026<\/span><\/div>\n<div class=\"inline-cta__register-button\">\n<div class=\"wp-block-buttons is-layout-flex wp-block-buttons-is-layout-flex\">\n<div class=\"wp-block-button\">REGISTER NOW<\/div>\n<\/div>\n<\/div>\n<\/div>\n<\/div>\n<\/div>\n<p class=\"wp-block-paragraph\">Notably, DeepSeek V4 is much more affordable than any frontier model available today. The smaller V4 Flash model costs $0.14 per million input tokens and $0.28 per million output tokens, undercutting GPT-5.4 Nano, Gemini 3.1 Flash, GPT-5.4 Mini, and Claude Haiku 4.5. The larger V4 Pro model, meanwhile, costs $0.145 per million input tokens and $3.48 per million output tokens, also undercutting Gemini 3.1 Pro, GPT-5.5, Claude Opus 4.7, and GPT-5.4.<\/p>\n<p class=\"wp-block-paragraph\">The launch comes a day after the U.S. accused China of stealing American AI labs\u2019 IP on an industrial scale using thousands of proxy accounts. DeepSeek itself has been accused by Anthropic and OpenAI of \u201cdistilling,\u201d essentially copying, their AI models.<\/p>\n<\/div>\n<p>[analyse_source url=&#8221;https:\/\/techcrunch.com\/2026\/04\/24\/deepseek-previews-new-ai-model-that-closes-the-gap-with-frontier-models\/&#8221;]<\/p>\n","protected":false},"excerpt":{"rendered":"<p>[analyse_image type=&#8221;featured&#8221; src=&#8221;https:\/\/techcrunch.com\/wp-content\/uploads\/2026\/04\/GettyImages-2272184430.jpg?w=1024&#8243;] Chinese AI lab DeepSeek has launched two preview versions of its newest large language model, DeepSeek V4, a much-awaited update to last year\u2019s V3.2 model and the accompanying R1 reasoning model that took the AI world by storm. The company says both DeepSeek V4 Flash and V4 Pro are mixture-of-experts models with [&hellip;]<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[2],"tags":[226,62],"class_list":["post-1901308","post","type-post","status-publish","format-standard","hentry","category-politics","tag-crawlmanager","tag-techcrunch-com"],"_links":{"self":[{"href":"https:\/\/analyse.optim.biz\/index.php?rest_route=\/wp\/v2\/posts\/1901308","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/analyse.optim.biz\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/analyse.optim.biz\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/analyse.optim.biz\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/analyse.optim.biz\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=1901308"}],"version-history":[{"count":0,"href":"https:\/\/analyse.optim.biz\/index.php?rest_route=\/wp\/v2\/posts\/1901308\/revisions"}],"wp:attachment":[{"href":"https:\/\/analyse.optim.biz\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=1901308"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/analyse.optim.biz\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=1901308"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/analyse.optim.biz\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=1901308"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}