{"id":2835,"date":"2024-07-19T10:39:51","date_gmt":"2024-07-19T10:39:51","guid":{"rendered":"https:\/\/howtogeek.blog\/cs\/?p=2835"},"modified":"2024-07-19T10:39:51","modified_gmt":"2024-07-19T10:39:51","slug":"openai-unveils-gpt-4o-mini-its-cheapest-small-ai-model-yet-cs","status":"publish","type":"post","link":"https:\/\/howtogeek.blog\/cs\/openai-unveils-gpt-4o-mini-its-cheapest-small-ai-model-yet-cs\/","title":{"rendered":"OpenAI odhaluje GPT-4o Mini, je to zat\u00edm nejlevn\u011bj\u0161\u00ed mal\u00fd model s um\u011blou inteligenc\u00ed"},"content":{"rendered":"

Dal\u0161\u00ed model um\u011bl\u00e9 inteligence se p\u0159ipojil k mo\u0159i model\u016f um\u011bl\u00e9 inteligence a tento je op\u011bt od OpenAI. OpenAI ji\u017e letos vydala GPT-4o, kter\u00e1 m\u011bla b\u00fdt efektivn\u011bj\u0161\u00ed ne\u017e GPT-4. St\u00e1le to v\u0161ak bylo pom\u011brn\u011b drah\u00e9 a mohlo to vyd\u011blat docela pen\u00edze, zejm\u00e9na pro v\u00fdvoj\u00e1\u0159e, kte\u0159\u00ed pot\u0159ebuj\u00ed volat model AI p\u0159es API pro sv\u00e9 aplikace opakovan\u011b b\u011bhem dne.<\/p>\n

V d\u016fsledku toho se v\u00fdvoj\u00e1\u0159i obr\u00e1tili na levn\u011bj\u0161\u00ed mal\u00e9 modely AI od konkurent\u016f, jako je Gemini 1.5 Flash nebo Claude 3 Haiku.<\/p>\n

Nyn\u00ed OpenAI uv\u00e1d\u00ed na trh GPT-4o mini, jejich dosud nejhospod\u00e1rn\u011bj\u0161\u00ed model, se kter\u00fdm tak\u00e9 vstupuje do mal\u00e9ho modelov\u00e9ho prostoru AI. Zat\u00edmco GPT-4o mini je jejich nejlevn\u011bj\u0161\u00ed model, nedosahuje tak n\u00edzk\u00fdch n\u00e1klad\u016f sn\u00ed\u017een\u00edm inteligence; je chyt\u0159ej\u0161\u00ed ne\u017e jejich st\u00e1vaj\u00edc\u00ed model GPT-3.5 Turbo.<\/p>\n

Podle OpenAI dos\u00e1hl GPT-4o mini 82 % v MMLU (Measuring Massive Multitask Language Understanding), \u010d\u00edm\u017e p\u0159ekonal mnoho model\u016f; toto jsou p\u0159\u00edslu\u0161n\u00e1 sk\u00f3re r\u016fzn\u00fdch model\u016f: GPT-3.5 Turbo (70 %), Claude 3 Haiku (75,2 %) a Gemini 1.5 Flash (78,9 %). GPT-4o dos\u00e1hl v tomto benchmarku 88,7 %, p\u0159i\u010dem\u017e Gemini Ultra se m\u016f\u017ee pochlubit nejvy\u0161\u0161\u00edm sk\u00f3re \u2013 90 % (nejedn\u00e1 se v\u0161ak o mal\u00e9 modely AI).<\/p>\n

\"\"
Zdroj: OpenAI<\/figcaption><\/figure>\n

GPT-4o mini je dnes k dispozici u\u017eivatel\u016fm ChatGPT Free, Team a Plus a tak\u00e9 v\u00fdvoj\u00e1\u0159\u016fm. Pro u\u017eivatele ChatGPT v podstat\u011b nahradil GPT-3.5; GPT-4o mini bude model, na kter\u00fd se konverzace jako v\u00fdchoz\u00ed pou\u017eije, jakmile v\u00e1m dojdou bezplatn\u00e9 dotazy GPT-4o. V\u00fdvoj\u00e1\u0159i budou m\u00edt st\u00e1le mo\u017enost pou\u017e\u00edvat GPT-3.5 prost\u0159ednictv\u00edm API, ale nakonec bude zru\u0161eno. U\u017eivatel\u00e9 ChatGPT Enterprise z\u00edskaj\u00ed p\u0159\u00edstup ke GPT-4o mini p\u0159\u00ed\u0161t\u00ed t\u00fdden.<\/p>\n

Jak bylo uvedeno v\u00fd\u0161e, GPT-4o mini se zam\u011b\u0159uje na pomoc v\u00fdvoj\u00e1\u0159\u016fm naj\u00edt pro jejich aplikaci model s n\u00edzkou cenou a latenc\u00ed, kter\u00fd je tak\u00e9 schopn\u00fd. Ve srovn\u00e1n\u00ed s jin\u00fdmi mal\u00fdmi modely vynik\u00e1 GPT-4o mini v \u00faloh\u00e1ch uva\u017eov\u00e1n\u00ed v textu i vid\u011bn\u00ed, matematick\u00e9m uva\u017eov\u00e1n\u00ed a k\u00f3dov\u00e1n\u00ed a tak\u00e9 v multimod\u00e1ln\u00edm uva\u017eov\u00e1n\u00ed.<\/p>\n

V sou\u010dasn\u00e9 dob\u011b podporuje jak text, tak vizi v API, s podporou textov\u00fdch, obrazov\u00fdch, obrazov\u00fdch a zvukov\u00fdch vstup\u016f a v\u00fdstup\u016f na pl\u00e1nu budoucnosti.<\/p>\n

GPT-4o mini m\u00e1 kontextov\u00e9 okno 128 000 token\u016f pro vstup a 16 000 token\u016f pro v\u00fdstup na po\u017eadavek, p\u0159i\u010dem\u017e jeho znalosti sahaj\u00ed a\u017e do \u0159\u00edjna 2023. Dok\u00e1\u017ee tak\u00e9 zpracovat neanglick\u00fd text pom\u011brn\u011b levn\u011b.<\/p>\n","protected":false},"excerpt":{"rendered":"

Dal\u0161\u00ed model um\u011bl\u00e9 inteligence se p\u0159ipojil k mo\u0159i model\u016f um\u011bl\u00e9 inteligence a tento je op\u011bt od OpenAI. OpenAI ji\u017e letos vydala GPT-4o, kter\u00e1 m\u011bla b\u00fdt efektivn\u011bj\u0161\u00ed ne\u017e GPT-4. St\u00e1le to v\u0161ak bylo pom\u011brn\u011b drah\u00e9 a mohlo to vyd\u011blat docela pen\u00edze, zejm\u00e9na pro v\u00fdvoj\u00e1\u0159e, kte\u0159\u00ed pot\u0159ebuj\u00ed volat model AI p\u0159es API pro sv\u00e9 aplikace opakovan\u011b b\u011bhem […]<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[1],"tags":[179,180,126],"class_list":["post-2835","post","type-post","status-publish","format-standard","hentry","category-how-to","tag-artificial-intelligence","tag-chatgpt","tag-microsoft"],"acf":[],"_links":{"self":[{"href":"https:\/\/howtogeek.blog\/cs\/wp-json\/wp\/v2\/posts\/2835","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/howtogeek.blog\/cs\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/howtogeek.blog\/cs\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/howtogeek.blog\/cs\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/howtogeek.blog\/cs\/wp-json\/wp\/v2\/comments?post=2835"}],"version-history":[{"count":1,"href":"https:\/\/howtogeek.blog\/cs\/wp-json\/wp\/v2\/posts\/2835\/revisions"}],"predecessor-version":[{"id":2836,"href":"https:\/\/howtogeek.blog\/cs\/wp-json\/wp\/v2\/posts\/2835\/revisions\/2836"}],"wp:attachment":[{"href":"https:\/\/howtogeek.blog\/cs\/wp-json\/wp\/v2\/media?parent=2835"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/howtogeek.blog\/cs\/wp-json\/wp\/v2\/categories?post=2835"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/howtogeek.blog\/cs\/wp-json\/wp\/v2\/tags?post=2835"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}