<div class="raw-html-embed"><div style="text-align: center;">
<img src="https://static-asst.8338.hk/udhk-cms/jx2sAu44sHwSZKDDTIwH.jpg" style=" width: 100%;max-width: 600px; height: auto;">
</div>
</div>The recent launch of OpenAI's GPT-4o model, with its new image generation capabilities, has sparked a global craze. However, this feature has also overwhelmed OpenAI's servers, pushing them into a state of emergency. The function, which can transform text prompts or photos into high-resolution artwork, particularly its ability to convert images into the Ghibli style, has exploded on social media, attracting millions of user-generated anime-style images in just a few days.&nbsp;This phenomenon has brought OpenAI's computing infrastructure to the brink of collapse. OpenAI CEO Sam Altman publicly complained about server overload on X, urging users to reduce the frequency of image generation to ensure the team has time to rest. This situation stems from users discovering, after the new feature went live on Thursday, that they could transform any photo into a Ghibli-style image, leading to viral spread. Even the White House joined in the creation, publishing a meme in the Ghibli style mocking a drug trafficking suspect.&nbsp;<h3>Ghibli Style Sweeps the Globe, Overloading Servers</h3>&nbsp;To cope with the server overload, OpenAI has taken emergency measures, suspending image generation in some regions and implementing a limit of three images per hour in others. This has sparked dissatisfaction among some users, who complain that the restrictions are too arbitrary. OpenAI subsequently announced that free users will be limited to three image generations per day, but this restriction will only be implemented after the feature is officially released to the public.&nbsp;Despite the server pressure, OpenAI has also gained a large number of new users as a result. According to statistics, "Ghibligeddon" attracted over one million people to register for ChatGPT accounts in one hour on Monday, although the actual number of users subscribing to paid services has not been disclosed. Google Trends data shows that the term "ChatGPT" briefly surpassed "Google" after the model's launch.&nbsp;GPT-4o's image generation technology differs from previous AI drawing tools in that it uses an autoregressive method, generating images sequentially pixel by pixel. While this method can produce more precise details and text rendering, it also significantly increases the GPU computing load, akin to the difference between creating an oil painting and a mosaic. Each image request triggers a large number of complex mathematical operations, consuming a significant amount of power and generating high heat. Running at maximum computing capacity for extended periods will damage the hardware.&nbsp;This server crisis comes at a time when OpenAI's financial situation is facing challenges. Although Altman claims the company is doing well, OpenAI is still losing billions of dollars annually, with losses projected to reach $14 billion by 2026. The image generation feature's massive computing demands further exacerbate the company's resource burden. While the Ghibli-style meme craze is not a revolutionary application that can generate billions of dollars in profit, its impact on OpenAI's servers should not be underestimated.&nbsp;Currently, OpenAI is actively optimizing its infrastructure in hopes of restoring the full operation of the image generation function as soon as possible, but has not announced a specific timeline. Users eager to experience the Ghibli-style transformation will need to wait patiently.

<div class="raw-html-embed"><div style="text-align: center;">
<img src="https://static-asst.8338.hk/udhk-cms/jx2sAu44sHwSZKDDTIwH.jpg" style=" width: 100%;max-width: 600px; height: auto;">
</div>
</div>OpenAI近期推出的GPT-4o模型中，其全新图像生成功能引发全球热潮，然而，这项功能也让OpenAI的伺服器不堪负荷，陷入紧急状态。该功能能将文字提示或照片转换为高解析度艺术作品，尤其其将图片转换为吉卜力风格的能力，更是引爆社群媒体，短短数日内便涌入数百万张用户创作的动漫风格图片。&nbsp;此现象导致OpenAI的运算基础设施面临崩溃边缘。 OpenAI执行长Sam Altman在X上公开抱怨伺服器超载，呼吁用户减少图片生成频率，以确保团队能有时间休息。此情况源于周四新功能上线后，用户发现能将任何照片转换成吉卜力风格图片，引发病毒式传播，甚至白宫也加入创作行列，发布了一张以吉卜力风格嘲讽贩毒嫌犯的迷因图。&nbsp;<h3>吉卜力风格席卷全球 令伺服器超载</h3>&nbsp;为应对伺服器超载，OpenAI已采取紧急措施，部分地区已暂停图像生成功能，其他地区则实施每小时限制生成三张图片的措施。此举引发部分用户不满，抱怨限制过于随意。 OpenAI随后宣布，免费用户每日生成图片数量将限制为三张，但此限制将在功能正式公开发布后才实施。&nbsp;尽管伺服器面临压力，OpenAI却也因此获得大量新用户。据统计，「吉卜力灾难」(Ghibligeddon) 在周一一个小时内吸引超过一百万人注册ChatGPT帐号，但实际订阅付费服务的用户数量尚未公开。 Google趋势数据显示，「ChatGPT」一词在该模型推出后一度超越「Google」。&nbsp;GPT-4o的图像生成技术与以往AI绘图工具不同，其采用自回归方法，以像素为单位依序生成图片，此方法虽能产生更精确的细节和文字渲染，却也大幅增加GPU运算负担，如同油画创作与马赛克创作的差异。每个图片请求都会触发大量复杂的数学运算，消耗大量电力并产生高热，长期以最大运算量运行将损坏硬体。&nbsp;这项伺服器危机正值OpenAI财务状况面临挑战之际。尽管Altman宣称公司发展良好，但OpenAI每年仍亏损数十亿美元，预估2026年亏损将达140亿美元。图像生成功能的庞大运算需求，更进一步加重公司资源负担。虽然吉卜力风格的迷因热潮并非能创造数十亿美元利润的革命性应用，但其对OpenAI伺服器造成的冲击不容小觑。&nbsp;目前，OpenAI正积极优化基础设施，以期尽快恢复图像生成功能的完整运作，但尚未公布确切时间。 渴望体验吉卜力风格转换的用户，则需耐心等待。

<div class="raw-html-embed"><div style="text-align: center;">
<img src="https://static-asst.8338.hk/udhk-cms/jx2sAu44sHwSZKDDTIwH.jpg" style=" width: 100%;max-width: 600px; height: auto;">
</div>
</div><a href="https://cftime.io/gpt-5-roadmap/">OpenAI</a>近期推出的GPT-4o模型中，其全新圖像生成功能引發全球熱潮，然而，這項功能也讓OpenAI的伺服器不堪負荷，陷入緊急狀態。該功能能將文字提示或照片轉換為高解析度藝術作品，尤其其將圖片轉換為吉卜力風格的能力，更是引爆社群媒體，短短數日內便湧入數百萬張用戶創作的動漫風格圖片。&nbsp;此現象導致OpenAI的運算基礎設施面臨崩潰邊緣。OpenAI執行長<a href="https://x.com/sama/status/1906210479695126886?ref_src=twsrc%5Etfw%7Ctwcamp%5Etweetembed%7Ctwterm%5E1906210479695126886%7Ctwgr%5E55419034e362246c669b8e88b074c1000e764d7c%7Ctwcon%5Es1_&amp;ref_url=https%3A%2F%2Fdecrypt.co%2F312482%2Fpeople-melting-openai-gpus-studio-ghibli-memes">Sam Altman在X上公開抱怨伺服器超載</a>，呼籲用戶減少圖片生成頻率，以確保團隊能有時間休息。此情況源於週四新功能上線後，用戶發現能將任何照片轉換成吉卜力風格圖片，引發病毒式傳播，甚至<a href="https://x.com/WhiteHouse/status/1905332049021415862">白宮也加入創作行列</a>，發布了一張以吉卜力風格嘲諷販毒嫌犯的迷因圖。&nbsp;<h3>吉卜力風格席捲全球 令伺服器超載</h3>&nbsp;為應對伺服器超載，OpenAI已採取緊急措施，部分地區已暫停圖像生成功能，其他地區則實施每小時限制生成三張圖片的措施。此舉引發部分用戶不滿，抱怨限制過於隨意。OpenAI隨後宣布，免費用戶每日生成圖片數量將限制為三張，但此限制將在功能正式公開發布後才實施。&nbsp;儘管伺服器面臨壓力，OpenAI卻也因此獲得大量新用戶。據統計，「吉卜力災難」(Ghibligeddon) 在週一一個小時內吸引超過一百萬人註冊ChatGPT帳號，但實際訂閱付費服務的用戶數量尚未公開。Google趨勢數據顯示，「ChatGPT」一詞在該模型推出後一度超越「Google」。&nbsp;GPT-4o的圖像生成技術與以往AI繪圖工具不同，<a href="https://cdn.openai.com/11998be9-5319-4302-bfbf-1167e093f1fb/Native_Image_Generation_System_Card.pdf">其採用自迴歸方法</a>，以像素為單位依序生成圖片，此方法雖能產生更精確的細節和文字渲染，卻也大幅增加GPU運算負擔，如同油畫創作與馬賽克創作的差異。每個圖片請求都會觸發大量複雜的數學運算，消耗大量電力並產生高熱，長期以最大運算量運行將損壞硬體。&nbsp;這項伺服器危機正值OpenAI財務狀況面臨挑戰之際。儘管Altman宣稱公司發展良好，但OpenAI每年仍虧損數十億美元，預估2026年虧損將達140億美元。圖像生成功能的龐大運算需求，更進一步加重公司資源負擔。雖然吉卜力風格的迷因熱潮並非能創造數十億美元利潤的革命性應用，但其對OpenAI伺服器造成的衝擊不容小覷。&nbsp;目前，OpenAI正積極優化基礎設施，以期盡快恢復圖像生成功能的完整運作，但尚未公布確切時間。 渴望體驗吉卜力風格轉換的用戶，則需耐心等待。

Ghibli style sweeps the world, OpenAI image generation function becomes popular and causes server crisis

吉卜力风格席卷全球 OpenAI图像生成功能爆红引发伺服器危机

吉卜力風格席捲全球 OpenAI圖像生成功能爆紅引發伺服器危機

<div class="raw-html-embed"><div style="text-align: center;">
<img src="https://static-asst.8338.hk/udhk-cms/I7CdTWivgqcyuAJBxSP2.jpg" style=" width: 100%;max-width: 600px; height: auto;">
</div>
</div>&nbsp;DeepSeek AI's ultra-efficient R1 model caused a global tech stock market crash, evaporating nearly a trillion dollars in market capitalization and fundamentally altering the competitive landscape of the AI industry. To counter this challenge, OpenAI urgently released its o3-mini model last Friday to compete with DeepSeek's R1. DeepSeek's R1 model shocked the entire AI industry with its top-tier performance at a remarkably low computational cost.&nbsp;OpenAI announced in its official blog post: "We're introducing OpenAI o3-mini, the latest and most cost-effective model in our reasoning family, now available on ChatGPT and the API platform." "This powerful and fast model, previewed in December 2024, pushes the boundaries of small model performance… while maintaining the low cost and low latency of OpenAI o1-mini."&nbsp;To promote its new reasoning model series, OpenAI offered reasoning capabilities to users for free for the first time and increased the daily message limit for paid users from 50 to 150 (a threefold increase).&nbsp;<h2>The o3-mini Model: Relatively Weak in Creativity, but Stronger in Problem Solving</h2>&nbsp;Unlike GPT-4o and the GPT series of models, the "o" series AI models focus on reasoning tasks. They are relatively weak in creativity but have built-in chain-of-thought reasoning capabilities, making them better at solving complex problems, correcting erroneous analyses, and writing better-structured code. OpenAI primarily has two AI model series: Generative Pre-trained Transformer (GPT) and "Omni" (o).&nbsp;The GPT series is like the artists in the family, excelling at role-playing, conversation, creative writing, summarization, explanation, brainstorming, and chatting; while the "o" series is like the scientists in the family, less adept at storytelling but skilled in coding, solving mathematical equations, analyzing complex problems, step-by-step planning and reasoning, and comparative research paper analysis.&nbsp;The new o3-mini model comes in low, medium, and high versions. Users can choose different versions based on their needs to obtain more accurate answers, but developers will need to pay more "reasoning" fees (charged by token).&nbsp;In terms of efficiency, OpenAI o3-mini underperforms OpenAI o1-mini in general knowledge and multilingual chain of thought, but scores higher on other tasks such as coding and factuality. The medium and large versions of o3-mini surpass OpenAI o1-mini in all benchmark tests.&nbsp;DeepSeek's R1 model, with its extremely low computational power yet surpassing the performance of OpenAI's flagship models, triggered a tech stock sell-off, resulting in nearly a trillion dollars in market cap loss in the US market, with Nvidia alone losing $600 billion. Investors became concerned about the future demand for its high-priced AI chips.&nbsp;DeepSeek's success stems from its innovative approach to model architecture. Unlike US companies that tend to invest more in computational power, the DeepSeek team focused on optimizing the model's information processing workflow to improve efficiency. With the Chinese tech giant Alibaba launching the even more powerful Qwen2.5 Max model (based on the same underlying model as DeepSeek), competitive pressure is further intensifying, signaling the arrival of a new wave of Chinese AI innovation.<h2>&nbsp;</h2><h2>OpenAI o3-mini: 24% Faster Than Its Predecessor</h2>&nbsp;OpenAI o3-mini attempts to regain the lead. The new model is 24% faster than its predecessor, performing on par with or even exceeding the old model in key benchmark tests, while also being more cost-effective.&nbsp;In terms of pricing, OpenAI o3-mini is priced at $0.55 per million input tokens and $4.40 per million output tokens. While higher than DeepSeek R1's $0.14 and $2.19, it has narrowed the price gap with DeepSeek and significantly reduced the cost compared to OpenAI o1. This is likely a key factor in its success. OpenAI o3-mini is a closed-source model, while DeepSeek R1 is open-source, but for users willing to pay for use on managed servers, the attractiveness of o3-mini will depend on its application scenarios.&nbsp;In the AIME math problem benchmark test, the OpenAI o3-mini medium version scored 79.6, DeepSeek R1 scored 79.8, second only to the OpenAI o3-mini large version (87.3).&nbsp;In other benchmark tests, such as the GPQA score measuring capabilities in different scientific fields, DeepSeek R1 scored 71.5, o3-mini low version scored 70.6, and o3-mini high version scored 79.7; in the Codeforces coding task benchmark test, R1 was at the 96.3rd percentile, o3-mini low version at the 93rd percentile, and o3-mini high version at the 97th percentile. Therefore, the differences between the models vary depending on the task.&nbsp;<h2>OpenAI o3-mini vs. DeepSeek R1: A Test Comparison</h2>&nbsp;We conducted several tests to compare the performance of the two models. One test was a spy game based on the Github BIG-bench dataset to assess multi-step reasoning ability. OpenAI o3-mini performed poorly in this test, reaching the wrong conclusion and misidentifying the culprit, while DeepSeek R1 correctly identified the culprit.&nbsp;However, o3-mini performed well in logic language tasks that did not involve mathematics. For example, when asked to write five sentences ending with a specific word, o3-mini was able to understand the task, evaluate the results, and provide the correct answer, taking four seconds and correcting one wrong answer itself.&nbsp;In mathematics, o3-mini excelled, quickly solving some problems considered extremely difficult. For example, a complex problem that took DeepSeek R1 275 seconds to solve was completed by o3-mini in just 33 seconds.&nbsp;In conclusion, OpenAI's o3-mini model shows some competitiveness, but the challenge from DeepSeek R1 remains, and the competition between the two in the AI field will continue to heat up.

<div class="raw-html-embed"><div style="text-align: center;">
<img src="https://static-asst.8338.hk/udhk-cms/Qi6I1bxTrUmcmE5cWov0.jpg" style=" width: 100%;max-width: 600px; height: auto;">
</div>
</div>&nbsp;DeepSeek AI 推出的超高效 R1 模型，曾引发全球科技股市值蒸发近一万亿美元的震荡，并彻底改变了人工智能产业的竞争格局。为应对这一挑战，OpenAI 于上周五紧急发布了 o3-mini模型，对撼 DeepSeek 的 R1 模型。 DeepSeek 的 R1 模型以其低廉的计算成本却能达到顶尖效能，震惊了整个 AI 行业。&nbsp;OpenAI 在官方博客文章中宣布：「我们推出OpenAI o3-mini，这是我们推理系列中最新、最具成本效益的模型，现已于ChatGPT 和API 平台上架。」 「这款于2024 年12 月预览的强大且高速模型，突破了小型模型的效能极限……同时保持OpenAI o1-mini 的低成本和低延迟。」&nbsp;为推广其全新推理模型系列，OpenAI 首次免费向用户提供推理功能，并将付费用户的每日讯息上限从 50 条提升至 150 条（增幅三倍）。&nbsp;<h2>o3-mini模型 创造力相对较弱 但更擅长解决复杂问题</h2>&nbsp;与 GPT-4o 和 GPT 系列模型不同，「o」系列 AI 模型专注于推理任务。其创造力相对较弱，但内建思维链推理功能，使其更擅长解决复杂问题、修正错误分析，并编写结构更佳的程式码。 OpenAI 主要拥有两个 AI 模型系列：生成式预训练变换器 (GPT) 和「Omni」 (o)。&nbsp;GPT 系列如同家族中的艺术家，擅长角色扮演、对话、创意写作、摘要、解释、脑力激荡、聊天等；而「o」系列则如同家族中的科学家，虽然不擅长叙事，却精于编码、解算数学方程式、分析复杂问题、逐步规划推理过程、比较研究论文等。&nbsp;全新的 o3-mini 模型分为低、中、高三个版本。用户可根据需求选择不同版本，以换取更精准的答案，但相对地，开发者需要支付更多「推理」费用（按代币计费）。&nbsp;就效率而言，OpenAI o3-mini 在一般知识和多语言思维链方面的表现不及 OpenAI o1-mini，但在编码和事实性等其他任务上的得分则更高。 o3-mini 中型和大型版本则在所有基准测试中均超越 OpenAI o1-mini。&nbsp;DeepSeek R1 模型以极低的计算能力却能超越OpenAI 旗舰模型的表现，引发了科技股抛售潮，导致美国市场市值损失近一万亿美元，其中英伟达单独市值蒸发6000 亿美元，投资者对其高价AI 晶片的未来需求产生疑虑。&nbsp;DeepSeek 的成功源于其在模型架构上的创新方法。与美国公司倾向于投入更多计算能力不同，DeepSeek 团队致力于优化模型的资讯处理流程，从而提升效率。随着中国科技巨头阿里巴巴推出效能更强的 Qwen2.5 Max 模型（其基础模型与 DeepSeek 相同），竞争压力进一步升级，预示着中国 AI 创新浪潮的到来。&nbsp;<h2>OpenAI o3-mini 运行速度比前代提升 24%</h2>&nbsp;OpenAI o3-mini 则试图再次拉大差距。新模型运行速度比前代提升 24%，在关键基准测试中与旧模型不相上下甚至超越，同时运作成本更低。&nbsp;在定价方面，OpenAI o3-mini 的价格为每百万个输入代币0.55 美元，每百万个输出代币4.40 美元，虽然高于DeepSeek R1 的0.14 美元和2.19 美元，但已缩小了与DeepSeek 的价格差距，并大幅降低了与OpenAI o1 相比的成本。这可能是其成功的关键因素。 OpenAI o3-mini 为闭源模型，而 DeepSeek R1 则为开源，但对于愿意在托管伺服器上付费使用的用户而言，o3-mini 的吸引力将取决于其应用场景。&nbsp;在 AIME 数学问题基准测试中，OpenAI o3-mini 中型版本得分 79.6 分，DeepSeek R1 得分 79.8 分，仅次于 OpenAI o3-mini 大型版本 (87.3 分)。&nbsp;在其他基准测试中，例如衡量不同科学领域能力的GPQA 分数，DeepSeek R1 为71.5 分，o3-mini 低配版为70.6 分，o3-mini 高配版为79.7 分；在编码任务基准测试Codeforces 中，R1位于第96.3 百分位，o3-mini 低配版位于第93 百分位，o3-mini 高配版位于第97 百分位。因此，不同模型之间的差异取决于任务而有所不同。<h2>&nbsp;</h2><h2>OpenAI o3-mini 与 DeepSeek R1 的测试比较</h2>&nbsp;我们进行了多项测试以比较两款模型的表现。其中一项测试为基于 Github BIG-bench 数据集的间谍游戏，用以评估多步骤推理能力。 OpenAI o3-mini 在此测试中表现不佳，得出错误结论，将凶手误判，而 DeepSeek R1 则正确地判断凶手。&nbsp;然而，o3-mini 在不涉及数学的逻辑语言任务中表现良好。例如，要求其撰写五个以特定单词结尾的句子，o3-mini 能够理解任务、评估结果并提供正确答案，其思考时间为四秒，并自行修正了一个错误答案。&nbsp;在数学方面，o3-mini 表现出色，能够快速解决一些被认为极其困难的问题。例如，DeepSeek R1 需要 275 秒才能解决的复杂问题，o3-mini 仅需 33 秒即可完成。&nbsp;总而言之，OpenAI 的 o3-mini 模型展现出一定的竞争力，但 DeepSeek R1 的挑战依然存在，双方在 AI 领域的竞争将持续升温。

<div class="raw-html-embed"><div style="text-align: center;">
<img src="https://static-asst.8338.hk/udhk-cms/Qi6I1bxTrUmcmE5cWov0.jpg" style=" width: 100%;max-width: 600px; height: auto;">
</div>
</div>&nbsp;<a href="https://cftime.io/deepseek-hits-the-world/">DeepSeek</a> AI 推出的超高效 R1 模型，曾引發全球科技股市值蒸發近一萬億美元的震盪，並徹底改變了人工智能產業的競爭格局。為應對這一挑戰，OpenAI 於上週五緊急發佈了 o3-mini模型，對撼 DeepSeek 的 R1 模型。DeepSeek 的 R1 模型以其低廉的計算成本卻能達到頂尖效能，震驚了整個 AI 行業。&nbsp;<a href="https://openai.com/index/openai-o3-mini/">OpenAI 在官方博客文章中宣佈</a>：「我們推出 OpenAI o3-mini，這是我們推理系列中最新、最具成本效益的模型，現已於 ChatGPT 和 API 平台上架。」 「這款於 2024 年 12 月預覽的強大且高速模型，突破了小型模型的效能極限……同時保持 OpenAI o1-mini 的低成本和低延遲。」&nbsp;為推廣其全新推理模型系列，OpenAI 首次免費向用戶提供推理功能，並將付費用戶的每日訊息上限從 50 條提升至 150 條（增幅三倍）。&nbsp;<h2>o3-mini模型 創造力相對較弱 但更擅長解決複雜問題</h2>&nbsp;與 GPT-4o 和 GPT 系列模型不同，「o」系列 AI 模型<a href="https://www.cloudflare.com/zh-tw/learning/ai/inference-vs-training/">專注於推理任務</a>。其創造力相對較弱，但內建思維鏈推理功能，使其更擅長解決複雜問題、修正錯誤分析，並編寫結構更佳的程式碼。OpenAI 主要擁有兩個 AI 模型系列：生成式預訓練變換器 (GPT) 和「Omni」 (o)。&nbsp;GPT 系列如同家族中的藝術家，擅長角色扮演、對話、創意寫作、摘要、解釋、腦力激盪、聊天等；而「o」系列則如同家族中的科學家，雖然不擅長敘事，卻精於編碼、解算數學方程式、分析複雜問題、逐步規劃推理過程、比較研究論文等。&nbsp;全新的 o3-mini 模型分為低、中、高三個版本。用戶可根據需求選擇不同版本，以換取更精準的答案，但相對地，開發者需要支付更多「推理」費用（按代幣計費）。&nbsp;就效率而言，OpenAI o3-mini 在一般知識和多語言思維鏈方面的表現不及 OpenAI o1-mini，但在編碼和事實性等其他任務上的得分則更高。o3-mini 中型和大型版本則在所有基準測試中均超越 OpenAI o1-mini。&nbsp;DeepSeek R1 模型以極低的計算能力卻能超越 OpenAI 旗艦模型的表現，引發了科技股拋售潮，導致美國市場市值損失近一萬億美元，其中英偉達單獨市值蒸發 6000 億美元，投資者對其高價 AI 晶片的未來需求產生疑慮。&nbsp;DeepSeek 的成功源於其在模型架構上的創新方法。與美國公司傾向於投入更多計算能力不同，DeepSeek 團隊致力於優化模型的資訊處理流程，從而提升效率。隨著中國科技巨頭阿里巴巴推出效能更強的 Qwen2.5 Max 模型（其基礎模型與 DeepSeek 相同），競爭壓力進一步升級，預示著中國 AI 創新浪潮的到來。&nbsp;<h2>OpenAI o3-mini 運行速度比前代提升 24%</h2>&nbsp;OpenAI o3-mini 則試圖再次拉大差距。新模型運行速度比前代提升 24%，在關鍵基準測試中與舊模型不相上下甚至超越，同時運作成本更低。&nbsp;在定價方面，OpenAI o3-mini 的價格為每百萬個輸入代幣 0.55 美元，每百萬個輸出代幣 4.40 美元，雖然高於 DeepSeek R1 的 0.14 美元和 2.19 美元，但已縮小了與 DeepSeek 的價格差距，並大幅降低了與 OpenAI o1 相比的成本。這可能是其成功的關鍵因素。OpenAI o3-mini 為閉源模型，而 DeepSeek R1 則為開源，但對於願意在託管伺服器上付費使用的用戶而言，o3-mini 的吸引力將取決於其應用場景。&nbsp;在 AIME 數學問題基準測試中，OpenAI o3-mini 中型版本得分 79.6 分，DeepSeek R1 得分 79.8 分，僅次於 OpenAI o3-mini 大型版本 (87.3 分)。&nbsp;在其他基準測試中，例如衡量不同科學領域能力的 GPQA 分數，DeepSeek R1 為 71.5 分，o3-mini 低配版為 70.6 分，o3-mini 高配版為 79.7 分；在編碼任務基準測試 Codeforces 中，R1 位於第 96.3 百分位，o3-mini 低配版位於第 93 百分位，o3-mini 高配版位於第 97 百分位。因此，不同模型之間的差異取決於任務而有所不同。&nbsp;<h2>OpenAI o3-mini 與 DeepSeek R1 的測試比較</h2>&nbsp;我們進行了多項測試以比較兩款模型的表現。其中一項測試為基於 Github BIG-bench 數據集的間諜遊戲，用以評估多步驟推理能力。OpenAI o3-mini 在此測試中表現不佳，得出錯誤結論，將兇手誤判，而 DeepSeek R1 則正確地判斷兇手。&nbsp;然而，o3-mini 在不涉及數學的邏輯語言任務中表現良好。例如，要求其撰寫五個以特定單詞結尾的句子，o3-mini 能夠理解任務、評估結果並提供正確答案，其思考時間為四秒，並自行修正了一個錯誤答案。&nbsp;在數學方面，o3-mini 表現出色，能夠快速解決一些被認為極其困難的問題。例如，DeepSeek R1 需要 275 秒才能解決的複雜問題，o3-mini 僅需 33 秒即可完成。&nbsp;總而言之，OpenAI 的 o3-mini 模型展現出一定的競爭力，但 DeepSeek R1 的挑戰依然存在，雙方在 AI 領域的競爭將持續升溫。

UD Blog

专业文章及见解，揭示科技领域的一切

最新 - # OpenAI

欢迎来到 UD！随时关注我们的分析文章、消息更新、优惠推广和新闻稿。

UD Blockchain 通讯