OpenAI o3

• 1. overview (651 chars)

  OpenAI o3 is a new generation flagship inference model released by OpenAI on April 17, 2025. It belongs to the o series of models and is designed to improve the problem-solving capabilities of ChatGPT. o3 is the most intelligent model of the o series to date, setting new records in the fields of programming, mathematics, science and visual perception. For the first time, images are directly integrated into the thinking chain, opening up a new way of solving problems that integrates visual and textual reasoning. Compared to the lightweight version o4-mini, o3 is better suited for complex queries and capable of generating and critically evaluating novel hypotheses. In terms of pricing, o3 input fees are $10 per million tokens and output fees are $40 per million tokens.

• 2. Product background (571 chars)

  OpenAI is an American artificial intelligence research organization founded in 2015 by Elon Musk and others and headquartered in San Francisco. The company's core mission is general artificial intelligence (AGI), and it has successively released the GPT series of large language models and the o series of inference models. On December 21, 2024, OpenAI announced the news of o3 and o4-mini for the first time at the 12th day conference, which attracted the attention of the global technology circle. On April 17, 2025, the two models were officially released, marking a major breakthrough for OpenAI in reasoning capabilities, multi-modal interaction and cost optimization.

• 3. Product functions and usage experience (1739 chars)

  The core functions of the o3 model are mainly reflected in the following aspects. The first is the significant improvement in reasoning ability. o3 is the most intelligent model in the o series so far. The reasoning ability has been greatly improved. The longer the thinking time, the better the effect. The second is multi-modal reasoning capability, which directly integrates images into the thinking chain for the first time, opening up a new problem-solving method that integrates visual and text reasoning. The third is the ability to call tools. For the first time, it fully supports web search, file analysis, Python code execution, visual input deep reasoning and image generation. It can quickly generate reliable answers in the correct format, usually taking less than a minute. The fourth is the ability to understand images. You can directly call tools to process images, and operations such as cropping, rotating, and scaling are easy. Even if the image is blurred, inverted, or of poor quality, the model can accurately interpret it. The fifth is the personalized memory function, which supports the memory function and can understand the user's interests and hobbies and provide personalized answers. The sixth is the search verification capability, which can call the search engine multiple times to cross-verify the results.   In terms of performance, o3 set new records in multiple benchmark tests. The visual task accuracy was 87.5%, and the MathVista test score was 75.4%. External expert evaluation shows that Programming, Business Consulting and Creative Ideation have 20% lower critical error rates than o1, and are particularly suitable for complex queries, with the ability to generate and critically evaluate novel hypotheses.   o4-mini is specially optimized as a lightweight version and is more suitable for scenarios that require fast response. The AIME 2024 math test has an accuracy rate of 92.7%, and the AIME 2025 math test has an accuracy rate of 93.4%. Outperforms the o3-mini and is more efficient in non-STEM and data science tasks.

• 4. Pricing and business model (844 chars)

  The pricing of the o3 model adopts a tiered strategy, and the specific prices are as follows: the o3 input fee is US$10 per million tokens, and the output fee is US$40 per million tokens. o4-mini has an input fee of $1.10 per million tokens and an output fee of $4.40 per million tokens. The processing length of approximately 750,000 tokens exceeds that of the Lord of the Rings series.   In June 2025, OpenAI announced that the o3 API price would drop by 80%, further lowering the threshold for developers to use it. o3-Pro focuses on deep reasoning, leading performance, and has obvious advantages in the STEM field. After the price reduction, o3 impacted the market and provided differentiated services for different needs.   In terms of availability, ChatGPT Plus, Pro, Team users can use it out of the box. Enterprise and education users will gain access a week later. Free users can use o4-mini in Think mode with unchanged rate limits. Developers can access it through the Chat Completions API and Responses API.

• 5. User reviews (991 chars)

  Judging from the user feedback searched, the release of o3 has aroused widespread attention and discussion. In terms of positive reviews, users generally recognize o3’s breakthroughs in reasoning capabilities and multi-modal processing. In particular, the image thinking function is regarded as an innovation. Most users agree that o3 performs better on math and programming tasks than its predecessor. The price/performance ratio after the price reduction has been recognized by some users.   Negative feedback mainly focused on pricing. Despite the price reduction, the cost of using o3 is still high, which is not easy for individual developers and small teams. Some users pointed out that the cost of using o3-PRO is significantly higher than the standard version. The stability of API access is also a concern for users, and the access experience in China needs to be optimized.   In terms of usage scenarios, o3 is particularly suitable for complex query scenarios that require deep reasoning, such as programming development, mathematical research, scientific computing, etc. For light tasks that require fast response, the o4-mini is a more cost-effective choice.

• 6. Industry evaluation (645 chars)

  From an industry perspective, the release of o3 is regarded as an important milestone in the development of AI technology. The industry generally believes that o3 has achieved significant improvements in reasoning capabilities, especially leading the way in STEM fields. The introduction of the image thinking function is considered an important breakthrough in multi-modal AI. The release of o3-PRO marks another step forward in AI technology.   However, the high cost of use limits its popularity. Some people believe that o3-PRO not only brings more powerful functions and more accurate answers, but also exposes some areas that need to be improved. For the majority of users and developers, o3 is both a tool full of opportunities and challenges in cost control.

• 7. Controversies and Risks (470 chars)

  At the technical level, the interpretability of deep AI reasoning is still limited, and its reliability in key application scenarios needs further verification. The carbon footprint issue caused by high reasoning costs has also attracted the attention of some environmentalists.   At the commercial level, continued high-end pricing strategies may affect the speed of user adoption. Some users have expressed concerns about API price fluctuations. The increasingly competitive market landscape poses potential challenges to OpenAI’s pricing power.

• 8. Applicable groups and usage suggestions (365 chars)

  o3 is particularly suitable for the following groups of people: professional developers and researchers, users who require complex reasoning capabilities, enterprise-level application scenarios, and users who require multi-modal processing.   For ordinary users and cost-sensitive users, o4-mini is a more cost-effective choice. ChatGPT Plus subscription also provides access to o3, suitable for general usage scenarios.

• 9. Summarize (387 chars)

  OpenAI o3 is currently the most intelligent model in the o series, achieving significant breakthroughs in reasoning capabilities, multi-modal understanding and complex task processing. The image thinking function introduced for the first time opens a new paradigm of AI reasoning. The high cost of use is still the main obstacle, and it has improved after the price reduction. It is recommended to choose the appropriate model version based on actual needs.

No in-depth report yet

• 

  Logan_Adams369

  —

  图像思考功能太香了，直接把截图丢进去就能分析，效率直接拉满。

---

• 

  NFsa_n

  —

  o3 的推理能力确实强，特别是复杂数学题，思考过程比 o1 详细太多了。

---

• 

  Amanda_Collins_88

  —

  价格还是太贵了，API 调用一次的成本够我用 Claude 好几次。

---

• 

  PaulBell_66

  —

  实测编程能力确实强，代码生成的质量比前代高一个档次。

---

• 

  Judy_Johnson_X

  —

  视觉任务 87.5% 准确率不是吹的，亲测有效。

---

• 

  whitesnake738

  —

  思维链可视化这个功能对学生党太友好了，可以学习 AI 的推理过程。

---

• 

  SharonFloresJr

  —

  降价 80% 后性价比高多了，之前嫌贵的可以再试试。

---

• 

  Bobby_Hall_8839

  —

  o3-pro 出来后果断订阅，深度推理确实香。

---

• 

  PatriciaMendoza_77

  —

  国内访问不稳定，经常超时，体验一般。

---

• 

  LiamGray

  —

  比 Google Gemini 和 Claude 都强，推理能力独一档。

---

• 

  smallbird262

  —

  MathVista 测试 75.4% 这分数太顶了，视觉理解目前最强。

---

• 

  EvelynRivera_20231

  —

  编程错误率比 o1 低 20%，实测写代码确实更稳了。

---

• 

  DAOthinker29

  —

  免费版用户体验阉割太多，不如加钱上 Plus。

---

• 

  Evelyn.RogersX88

  —

  企业用户一周后才能用，等得好焦虑。

---

• 

  7EXK6NT8MN

  —

  75 万 tokens 处理量，约等于《指环王》三部曲，这上下文太离谱。

---

• 

  Rita816

  —

  ChatGPT Plus 订阅就能用，比单独买 API 划算。

---

• 

  任博然

  —

  o4-mini 性价比更高，普通任务完全够用。

---

• 

  Matthew_Patel_71

  —

  多模态融合是最大亮点，图像和文本一起推理的体验很新鲜。

---

• 

  organicfrog247

  —

  OpenAI 史上最强推理模型实至名归，虽然贵但确实强。

No reviews yet