{"id":777,"date":"2024-12-19T10:00:00","date_gmt":"2024-12-19T10:00:00","guid":{"rendered":"https:\/\/jacar.es\/o1-preview-razonamiento\/"},"modified":"2024-12-19T10:00:00","modified_gmt":"2024-12-19T10:00:00","slug":"o1-preview-razonamiento","status":"publish","type":"post","link":"https:\/\/jacar.es\/en\/o1-preview-razonamiento\/","title":{"rendered":"o1-preview: OpenAI’s Model That Thinks Before Answering"},"content":{"rendered":"

OpenAI introduced o1-preview<\/strong> and o1-mini<\/strong> on September 12, 2024 as a new model family with a philosophical difference from GPT-4o: they reason internally<\/strong> before emitting the final answer. It\u2019s not just visible chain-of-thought; it\u2019s a hidden process where the model elaborates, reconsiders, explores paths, and only then responds. Results in mathematics and code are dramatic. For other tasks, the latency and cost trade-off isn\u2019t always worth it.<\/p>\n

What Makes It Different<\/h2>\n

Traditional models (GPT-4o, Claude 3.5 Sonnet) generate response tokens sequentially from the first token. o1 introduces a prior user-invisible \u201cthinking\u201d phase where the model can reformulate the problem, explore approaches, discard strategies, before elaborating the answer. This internal reasoning consumes additional tokens \u2014 so-called reasoning tokens<\/em> \u2014 which are billed but not shown to the user.<\/p>\n

The approach is partially inspired by techniques like Tree of Thoughts and MCTS but integrated at training level, not prompt. OpenAI doesn\u2019t detail exact architecture; we know there\u2019s a reinforcement-learning phase teaching the model to reason efficiently on complex problems.<\/p>\n

Where It Excels<\/h2>\n

Benchmarks tell a clear story. On maths problems like AIME (American Invitational Mathematics Examination), o1-preview scores 83% vs GPT-4o\u2019s 13%. On PhD-level physics problems, o1 reaches 78% vs 57% of human experts. In programming competitions like Codeforces, o1-preview sits at 89th percentile vs GPT-4o\u2019s 11th.<\/p>\n

For tasks requiring long chains of reasoning \u2014 olympiad maths, proofs, complex debugging, deep causal analysis \u2014 the qualitative leap is real and not just statistical. Users report o1 solves problems where GPT-4o went in circles.<\/p>\n

Where It Doesn\u2019t Add Much<\/h2>\n

For conversational tasks, creative writing, simple summaries, direct factual questions, o1 offers no significant advantage over GPT-4o, while adding latency and cost. A response GPT-4o generates in one second may take ten to twenty in o1 while it \u201cthinks\u201d. For fluid conversation, that breaks the experience.<\/p>\n

Additionally, o1 has architectural limitations. It doesn\u2019t (at least in preview) support function calling the same way as GPT-4o, doesn\u2019t stream, isn\u2019t multimodal. For workflows depending on those capabilities, it\u2019s not a direct replacement.<\/p>\n

The Cost Factor<\/h2>\n

o1-preview costs $15 per million input tokens and $60 per million output. o1-mini is cheaper \u2014 $3 input, $12 output \u2014 and is typically the sweet spot for most uses benefiting from reasoning. For comparison, GPT-4o costs $2.50 and $10 respectively.<\/p>\n

Real cost is higher than nominal because hidden reasoning tokens consume output billing. An apparently short response may have internally consumed ten times the visible tokens. For high-volume applications, this adds up.<\/p>\n

When to Use and When Not<\/h2>\n

The pragmatic rule emerging after weeks of use: for problems where the answer requires several chained reasoning steps and where correctness matters more than speed, o1 is worth it. Complex technical research, legal analysis with multiple premises, maths or programming problems with rich structure, strategic planning with interdependent variables.<\/p>\n

For chatbots, content generation, summaries, translations, quick questions, function calling, o1 introduces cost and latency without proportional benefit. GPT-4o or Claude 3.5 Sonnet are better choices.<\/p>\n

The Industry Effect<\/h2>\n

o1 marked a conceptual shift. Previously, progress came primarily from scaling parameters and data (GPT-3 \u2192 GPT-4). o1 shows scaling inference-time compute \u2014 giving the model more tokens to think \u2014 also produces qualitative leaps. This opens a new scaling dimension.<\/p>\n

Shortly after, Anthropic announced similar capabilities in later Claude versions. Google prepares its response in the Gemini family. Reasoning-model competition is already active battleground. For 2025 we expect multiple options with similar paradigms.<\/p>\n

Limitations and Criticism<\/h2>\n

Being honest about problems. Internal reasoning isn\u2019t transparent \u2014 OpenAI explicitly hides reasoning tokens from users. This generates legitimate concerns about auditing and debugging. How to debug an error when you can\u2019t see the reasoning that produced it?<\/p>\n

Published benchmarks are somewhat cherry-picked. o1 isn\u2019t universally superior; on many everyday tasks it ties or loses to GPT-4o when normalised for cost. The \u201co1 is better at everything\u201d narrative is incorrect.<\/p>\n

There are also open questions about the approach\u2019s sustainability. If each model generation requires ten times more inferential tokens, carbon footprint and economic cost scale exponentially. At some point this growth stabilises or gets rethought.<\/p>\n

Practical Integration<\/h2>\n

For teams wanting to incorporate o1 into their stack, the pragmatic approach is multi-model routing: use GPT-4o or Claude for most queries and escalate to o1 only when the task justifies. Tools like LiteLLM facilitate this pattern with a unified proxy.<\/p>\n

Another useful pattern is \u201creviewer\u201d mode: o1 reviews responses produced by cheaper models and flags reasoning errors. Per-review cost is low because responses are already structured; quality benefit can be high for critical tasks.<\/p>\n

Conclusion<\/h2>\n

o1 represents an inflection point in how we think about language models. It\u2019s not a universal GPT-4o replacement but a specialised complement for deep reasoning. For problems where correctness matters more than speed, it\u2019s worth every extra cent. For most everyday uses, traditional models remain more efficient. The direction it marks \u2014 scaling inference compute for reasoning \u2014 is probably the next dominant paradigm. Knowing when to apply it forms part of essential technical repertoire for any engineer integrating LLMs in production.<\/p>\n

Follow us on jacar.es for more on reasoning models, frontier LLMs, and multi-model strategies.<\/p>\n","protected":false},"excerpt":{"rendered":"

OpenAI introduced o1 in September 2024. A model that reasons internally before answering. What changes and when it’s worth it.<\/p>\n","protected":false},"author":1,"featured_media":778,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[22],"tags":[129,724,722,51,723],"class_list":["post-777","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-inteligencia-artificial","tag-benchmarks","tag-chain-of-thought","tag-o1","tag-openai","tag-reasoning"],"translation":{"provider":"WPGlobus","version":"3.0.2","language":"en","enabled_languages":["es","en"],"languages":{"es":{"title":true,"content":true,"excerpt":true},"en":{"title":true,"content":true,"excerpt":true}}},"gutentor_comment":0,"yoast_head":"\no1-preview: OpenAI's Model That Thinks Before Answering - Jacar<\/title>\n<meta name=\"description\" content=\"OpenAI o1-preview: internal reasoning before answering, math benchmarks, when it beats GPT-4o, and limitations.\" \/>\n<meta name=\"robots\" content=\"index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1\" \/>\n<link rel=\"canonical\" href=\"https:\/\/jacar.es\/o1-preview-razonamiento\/\" \/>\n<meta property=\"og:locale\" content=\"en_US\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"o1-preview: OpenAI's Model That Thinks Before Answering - Jacar\" \/>\n<meta property=\"og:description\" content=\"OpenAI o1-preview: internal reasoning before answering, math benchmarks, when it beats GPT-4o, and limitations.\" \/>\n<meta property=\"og:url\" content=\"https:\/\/jacar.es\/o1-preview-razonamiento\/\" \/>\n<meta property=\"og:site_name\" content=\"Jacar\" \/>\n<meta property=\"article:published_time\" content=\"2024-12-19T10:00:00+00:00\" \/>\n<meta property=\"og:image\" content=\"https:\/\/jcs-wp-jacar-es.fsn1.your-objectstorage.com\/wp-content\/uploads\/2020\/09\/favicon.png\" \/>\n\t<meta property=\"og:image:width\" content=\"252\" \/>\n\t<meta property=\"og:image:height\" content=\"229\" \/>\n\t<meta property=\"og:image:type\" content=\"image\/png\" \/>\n<meta name=\"author\" content=\"javi\" \/>\n<meta name=\"twitter:card\" content=\"summary_large_image\" \/>\n<meta name=\"twitter:label1\" content=\"Written by\" \/>\n\t<meta name=\"twitter:data1\" content=\"javi\" \/>\n\t<meta name=\"twitter:label2\" content=\"Est. reading time\" \/>\n\t<meta name=\"twitter:data2\" content=\"10 minutes\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\\\/\\\/schema.org\",\"@graph\":[{\"@type\":\"Article\",\"@id\":\"https:\\\/\\\/jacar.es\\\/o1-preview-razonamiento\\\/#article\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/jacar.es\\\/o1-preview-razonamiento\\\/\"},\"author\":{\"name\":\"javi\",\"@id\":\"https:\\\/\\\/jacar.es\\\/#\\\/schema\\\/person\\\/54a7f7b4224b38fafc9866eb3e614208\"},\"headline\":\"o1-preview: OpenAI’s Model That Thinks Before Answering\",\"datePublished\":\"2024-12-19T10:00:00+00:00\",\"mainEntityOfPage\":{\"@id\":\"https:\\\/\\\/jacar.es\\\/o1-preview-razonamiento\\\/\"},\"wordCount\":1932,\"publisher\":{\"@id\":\"https:\\\/\\\/jacar.es\\\/#organization\"},\"image\":{\"@id\":\"https:\\\/\\\/jacar.es\\\/o1-preview-razonamiento\\\/#primaryimage\"},\"thumbnailUrl\":\"https:\\\/\\\/jcs-wp-jacar-es.fsn1.your-objectstorage.com\\\/wp-content\\\/uploads\\\/2024\\\/12\\\/20053704\\\/jwp-1871820-28642.jpg\",\"keywords\":[\"benchmarks\",\"chain of thought\",\"o1\",\"openai\",\"reasoning\"],\"articleSection\":[\"Inteligencia Artificial\"],\"inLanguage\":\"en-US\"},{\"@type\":[\"WebPage\",\"ItemPage\"],\"@id\":\"https:\\\/\\\/jacar.es\\\/o1-preview-razonamiento\\\/\",\"url\":\"https:\\\/\\\/jacar.es\\\/o1-preview-razonamiento\\\/\",\"name\":\"o1-preview: OpenAI's Model That Thinks Before Answering - Jacar\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/jacar.es\\\/#website\"},\"primaryImageOfPage\":{\"@id\":\"https:\\\/\\\/jacar.es\\\/o1-preview-razonamiento\\\/#primaryimage\"},\"image\":{\"@id\":\"https:\\\/\\\/jacar.es\\\/o1-preview-razonamiento\\\/#primaryimage\"},\"thumbnailUrl\":\"https:\\\/\\\/jcs-wp-jacar-es.fsn1.your-objectstorage.com\\\/wp-content\\\/uploads\\\/2024\\\/12\\\/20053704\\\/jwp-1871820-28642.jpg\",\"datePublished\":\"2024-12-19T10:00:00+00:00\",\"description\":\"OpenAI o1-preview: internal reasoning before answering, math benchmarks, when it beats GPT-4o, and limitations.\",\"breadcrumb\":{\"@id\":\"https:\\\/\\\/jacar.es\\\/o1-preview-razonamiento\\\/#breadcrumb\"},\"inLanguage\":\"en-US\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\\\/\\\/jacar.es\\\/o1-preview-razonamiento\\\/\"]}]},{\"@type\":\"ImageObject\",\"inLanguage\":\"en-US\",\"@id\":\"https:\\\/\\\/jacar.es\\\/o1-preview-razonamiento\\\/#primaryimage\",\"url\":\"https:\\\/\\\/jcs-wp-jacar-es.fsn1.your-objectstorage.com\\\/wp-content\\\/uploads\\\/2024\\\/12\\\/20053704\\\/jwp-1871820-28642.jpg\",\"contentUrl\":\"https:\\\/\\\/jcs-wp-jacar-es.fsn1.your-objectstorage.com\\\/wp-content\\\/uploads\\\/2024\\\/12\\\/20053704\\\/jwp-1871820-28642.jpg\",\"width\":1200,\"height\":800,\"caption\":\"Monta\u00f1as al amanecer con paisaje reflexivo representando contemplaci\u00f3n profunda\"},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\\\/\\\/jacar.es\\\/o1-preview-razonamiento\\\/#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"Portada\",\"item\":\"https:\\\/\\\/jacar.es\\\/\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"o1-preview: el modelo de OpenAI que piensa antes de responder\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\\\/\\\/jacar.es\\\/#website\",\"url\":\"https:\\\/\\\/jacar.es\\\/\",\"name\":\"Jacar\",\"description\":\"Passion for Technology\",\"publisher\":{\"@id\":\"https:\\\/\\\/jacar.es\\\/#organization\"},\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\\\/\\\/jacar.es\\\/?s={search_term_string}\"},\"query-input\":{\"@type\":\"PropertyValueSpecification\",\"valueRequired\":true,\"valueName\":\"search_term_string\"}}],\"inLanguage\":\"en-US\"},{\"@type\":\"Organization\",\"@id\":\"https:\\\/\\\/jacar.es\\\/#organization\",\"name\":\"Jacar\",\"url\":\"https:\\\/\\\/jacar.es\\\/\",\"logo\":{\"@type\":\"ImageObject\",\"inLanguage\":\"en-US\",\"@id\":\"https:\\\/\\\/jacar.es\\\/#\\\/schema\\\/logo\\\/image\\\/\",\"url\":\"https:\\\/\\\/jacar.es\\\/wp-content\\\/uploads\\\/2020\\\/09\\\/favicon.png\",\"contentUrl\":\"https:\\\/\\\/jacar.es\\\/wp-content\\\/uploads\\\/2020\\\/09\\\/favicon.png\",\"width\":252,\"height\":229,\"caption\":\"Jacar\"},\"image\":{\"@id\":\"https:\\\/\\\/jacar.es\\\/#\\\/schema\\\/logo\\\/image\\\/\"},\"sameAs\":[\"https:\\\/\\\/www.linkedin.com\\\/in\\\/javiercanetearroyo\\\/\"]},{\"@type\":\"Person\",\"@id\":\"https:\\\/\\\/jacar.es\\\/#\\\/schema\\\/person\\\/54a7f7b4224b38fafc9866eb3e614208\",\"name\":\"javi\",\"sameAs\":[\"https:\\\/\\\/jacar.es\"],\"url\":\"https:\\\/\\\/jacar.es\\\/en\\\/author\\\/javi\\\/\"}]}<\/script>\n<!-- \/ Yoast SEO plugin. -->","yoast_head_json":{"title":"o1-preview: OpenAI's Model That Thinks Before Answering - Jacar","description":"OpenAI o1-preview: internal reasoning before answering, math benchmarks, when it beats GPT-4o, and limitations.","robots":{"index":"index","follow":"follow","max-snippet":"max-snippet:-1","max-image-preview":"max-image-preview:large","max-video-preview":"max-video-preview:-1"},"canonical":"https:\/\/jacar.es\/o1-preview-razonamiento\/","og_locale":"en_US","og_type":"article","og_title":"o1-preview: OpenAI's Model That Thinks Before Answering - Jacar","og_description":"OpenAI o1-preview: internal reasoning before answering, math benchmarks, when it beats GPT-4o, and limitations.","og_url":"https:\/\/jacar.es\/o1-preview-razonamiento\/","og_site_name":"Jacar","article_published_time":"2024-12-19T10:00:00+00:00","og_image":[{"width":252,"height":229,"url":"https:\/\/jcs-wp-jacar-es.fsn1.your-objectstorage.com\/wp-content\/uploads\/2020\/09\/favicon.png","type":"image\/png"}],"author":"javi","twitter_card":"summary_large_image","twitter_misc":{"Written by":"javi","Est. reading time":"10 minutes"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"Article","@id":"https:\/\/jacar.es\/o1-preview-razonamiento\/#article","isPartOf":{"@id":"https:\/\/jacar.es\/o1-preview-razonamiento\/"},"author":{"name":"javi","@id":"https:\/\/jacar.es\/#\/schema\/person\/54a7f7b4224b38fafc9866eb3e614208"},"headline":"o1-preview: OpenAI’s Model That Thinks Before Answering","datePublished":"2024-12-19T10:00:00+00:00","mainEntityOfPage":{"@id":"https:\/\/jacar.es\/o1-preview-razonamiento\/"},"wordCount":1932,"publisher":{"@id":"https:\/\/jacar.es\/#organization"},"image":{"@id":"https:\/\/jacar.es\/o1-preview-razonamiento\/#primaryimage"},"thumbnailUrl":"https:\/\/jcs-wp-jacar-es.fsn1.your-objectstorage.com\/wp-content\/uploads\/2024\/12\/20053704\/jwp-1871820-28642.jpg","keywords":["benchmarks","chain of thought","o1","openai","reasoning"],"articleSection":["Inteligencia Artificial"],"inLanguage":"en-US"},{"@type":["WebPage","ItemPage"],"@id":"https:\/\/jacar.es\/o1-preview-razonamiento\/","url":"https:\/\/jacar.es\/o1-preview-razonamiento\/","name":"o1-preview: OpenAI's Model That Thinks Before Answering - Jacar","isPartOf":{"@id":"https:\/\/jacar.es\/#website"},"primaryImageOfPage":{"@id":"https:\/\/jacar.es\/o1-preview-razonamiento\/#primaryimage"},"image":{"@id":"https:\/\/jacar.es\/o1-preview-razonamiento\/#primaryimage"},"thumbnailUrl":"https:\/\/jcs-wp-jacar-es.fsn1.your-objectstorage.com\/wp-content\/uploads\/2024\/12\/20053704\/jwp-1871820-28642.jpg","datePublished":"2024-12-19T10:00:00+00:00","description":"OpenAI o1-preview: internal reasoning before answering, math benchmarks, when it beats GPT-4o, and limitations.","breadcrumb":{"@id":"https:\/\/jacar.es\/o1-preview-razonamiento\/#breadcrumb"},"inLanguage":"en-US","potentialAction":[{"@type":"ReadAction","target":["https:\/\/jacar.es\/o1-preview-razonamiento\/"]}]},{"@type":"ImageObject","inLanguage":"en-US","@id":"https:\/\/jacar.es\/o1-preview-razonamiento\/#primaryimage","url":"https:\/\/jcs-wp-jacar-es.fsn1.your-objectstorage.com\/wp-content\/uploads\/2024\/12\/20053704\/jwp-1871820-28642.jpg","contentUrl":"https:\/\/jcs-wp-jacar-es.fsn1.your-objectstorage.com\/wp-content\/uploads\/2024\/12\/20053704\/jwp-1871820-28642.jpg","width":1200,"height":800,"caption":"Monta\u00f1as al amanecer con paisaje reflexivo representando contemplaci\u00f3n profunda"},{"@type":"BreadcrumbList","@id":"https:\/\/jacar.es\/o1-preview-razonamiento\/#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"Portada","item":"https:\/\/jacar.es\/"},{"@type":"ListItem","position":2,"name":"o1-preview: el modelo de OpenAI que piensa antes de responder"}]},{"@type":"WebSite","@id":"https:\/\/jacar.es\/#website","url":"https:\/\/jacar.es\/","name":"Jacar","description":"Passion for Technology","publisher":{"@id":"https:\/\/jacar.es\/#organization"},"potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/jacar.es\/?s={search_term_string}"},"query-input":{"@type":"PropertyValueSpecification","valueRequired":true,"valueName":"search_term_string"}}],"inLanguage":"en-US"},{"@type":"Organization","@id":"https:\/\/jacar.es\/#organization","name":"Jacar","url":"https:\/\/jacar.es\/","logo":{"@type":"ImageObject","inLanguage":"en-US","@id":"https:\/\/jacar.es\/#\/schema\/logo\/image\/","url":"https:\/\/jacar.es\/wp-content\/uploads\/2020\/09\/favicon.png","contentUrl":"https:\/\/jacar.es\/wp-content\/uploads\/2020\/09\/favicon.png","width":252,"height":229,"caption":"Jacar"},"image":{"@id":"https:\/\/jacar.es\/#\/schema\/logo\/image\/"},"sameAs":["https:\/\/www.linkedin.com\/in\/javiercanetearroyo\/"]},{"@type":"Person","@id":"https:\/\/jacar.es\/#\/schema\/person\/54a7f7b4224b38fafc9866eb3e614208","name":"javi","sameAs":["https:\/\/jacar.es"],"url":"https:\/\/jacar.es\/en\/author\/javi\/"}]}},"_links":{"self":[{"href":"https:\/\/jacar.es\/en\/wp-json\/wp\/v2\/posts\/777","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/jacar.es\/en\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/jacar.es\/en\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/jacar.es\/en\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/jacar.es\/en\/wp-json\/wp\/v2\/comments?post=777"}],"version-history":[{"count":0,"href":"https:\/\/jacar.es\/en\/wp-json\/wp\/v2\/posts\/777\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/jacar.es\/en\/wp-json\/wp\/v2\/media\/778"}],"wp:attachment":[{"href":"https:\/\/jacar.es\/en\/wp-json\/wp\/v2\/media?parent=777"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/jacar.es\/en\/wp-json\/wp\/v2\/categories?post=777"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/jacar.es\/en\/wp-json\/wp\/v2\/tags?post=777"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}