OpenAI Unveils Enhanced Transparency in o3-mini Model Amidst Growing Competition

OpenAI's commitment to enhancing the transparency of its AI models marks a significant step in the evolution of artificial intelligence technology. This initiative specifically focuses on the o3‑mini models, aiming to reveal more of their reasoning processes. In response to competitive pressures from companies like DeepSeek, OpenAI is motivated to not only enhance its model's capabilities but also to address growing user demand for clearer insights into how AI makes decisions. By offering more detailed summaries of the AI's thought process and integrating features such as multi‑language support, OpenAI strives to meet these expectations and simultaneously maintain competitive leverage in the technology sector.

The core changes introduced in OpenAI's o3‑mini models include comprehensive summaries that elucidate the reasoning steps undertaken by the AI. This effort is not just about transparency but also about improving safety and usability. The models now incorporate a post‑processing step designed to remove unsafe content, ensuring that the AI's outputs are not only transparent but also safe for widespread consumption. Moreover, the introduction of multi‑language summaries broadens the accessibility of these features to a global audience, thus fostering inclusivity and broader understanding of AI technologies.

Despite these advancements, OpenAI's transparency initiative does not fully disclose the complete raw chain of thought. The decision to withhold certain aspects of the model's reasoning is reflective of a strategic balance between offering significant transparency and safeguarding proprietary technologies. This approach underscores the trade‑offs that often accompany transparency in AI development, as companies must weigh the benefits of openness against the risks of exposing sensitive data to competitors.

The public's reception of OpenAI's transparency measures in the o3‑mini model has been largely positive, with many users appreciating the enhanced clarity in AI decision‑making processes. The decision to extend these transparency features to both free and paid users further underscores OpenAI’s commitment to democratizing AI technology. However, some critiques focus on the performance aspects of the model, particularly during peak usage times. There is also ongoing discussion comparing OpenAI’s strategy with competitors like DeepSeek, who favor a more open‑source approach. Nevertheless, the initiative is a pivotal move towards more responsible and transparent AI development.

In today's rapidly evolving technological landscape, transparency has become a key driver for innovation and trust, especially in the realm of artificial intelligence. Companies like OpenAI are feeling increased competitive pressure from innovators like DeepSeek to provide users with deeper insights into the workings of their AI models. This move towards transparency is not just about staying competitive; it addresses a growing user demand for clearer understanding of AI reasoning processes. By revealing more about the thought processes of their models, companies are aiming to correct past issues related to inaccurate summaries, thereby improving user trust and satisfaction..¹

Moreover, the drive for transparency is fueled by an imperative to maintain ethical practices and responsibility in AI deployments. Understanding how AI makes decisions is crucial to ensuring that these systems do not perpetuate biases or generate unsafe content. OpenAI's o3‑mini model, for example, has introduced a post‑processing step specifically designed to weed out unsafe content, aligning with broader industry calls for responsible AI practices. These advancements not only aim to empower users with more accurate information but also protect them by mitigating potential risks associated with AI misjudgment.¹

Transparency initiatives like those undertaken by OpenAI also serve a broader societal and regulatory function. By providing detailed reasoning summaries, these models invite scrutiny and dialogue about AI capabilities, promoting public understanding and trust. These efforts are echoed in global regulatory environments, as seen in initiatives such as the EU AI Safety Coalition, which demands detailed documentation of AI decision‑making processes. Such transparency is increasingly seen as foundational, not just for user engagement, but also for overcoming legal and ethical hurdles in AI deployment. For a detailed overview of such regulatory initiatives, click here.

OpenAI's decision to enhance the transparency of its o3‑mini model is a crucial step forward in the ongoing quest for accountability and clarity in AI development. As competitive pressures mount, particularly from formidable rivals like DeepSeek, OpenAI is working to maintain its edge by responding to user demands for a clearer understanding of how models arrive at their conclusions. This need for transparency is underscored by past challenges where users encountered inaccuracies in generated summaries. To address these issues, OpenAI has implemented more detailed reasoning summaries and innovative post‑processing steps to eliminate unsafe content. These changes not only simplify complex AI‑generated ideas but also extend support to non‑English speakers, thus broadening the reach and applicability of the o3‑mini model.¹

The redesigned o3‑mini model represents a significant evolution in AI reasoning technology, a field that has previously been riddled with simple errors due to a lack of self‑verification mechanisms in simpler models. The incorporation of self‑fact‑checking capabilities enables the model to reduce these common errors by verifying the accuracy of its outputs before delivering results, albeit with the trade‑off of slightly slower response times. This advancement sets a new standard in AI development, ensuring that reasoning models are not merely reactive but proactive in error detection.¹

Despite these improvements, OpenAI has taken a measured approach by not fully exposing the raw chain of thought of its models. This careful balance between enhanced transparency and the protection of proprietary information reflects competitive concerns that keep OpenAI ahead of its industry peers while cautiously extending its innovations to all ChatGPT users. Both free and paid users can leverage these improvements through the o3‑mini and o3‑mini‑high configurations, thereby democratizing access to advanced AI technology.¹

The reception of these changes has been split across various spheres of public opinion. Many users and industry experts like Dr. Sarah Chen applaud these efforts, citing increased transparency as a key to boosting user trust and facilitating practical applications in research and education. However, some experts, such as Dr. Elena Rodriguez, caution that the reliance on post‑processing filters could inadvertently create "blind spots," potentially undermining the intended transparency. OpenAI's acknowledgement of the filtered nature of their summaries further feeds into this nuanced debate, as they strive to strike a balance between disclosure and intellectual property protection.¹

Looking ahead, the enhancements introduced with the o3‑mini model are bound to have a ripple effect on various aspects of the AI landscape. Economically, increased trust in AI technologies may lead to broader adoption across different industries, significantly driving productivity gains. However, the push for more transparent models also necessitates higher development costs as companies strive to bolster their explainability and auditing mechanisms. These financial pressures may favor larger firms that have the resources to meet these transparency demands .

The recent shift towards increased transparency in AI models, spearheaded by OpenAI's latest initiatives, aims to unveil more detailed insights into how AI systems generate their responses. OpenAI's o3‑mini model is at the forefront of this movement, offering enhanced reasoning summaries that aim to elucidate the underlying thought processes of AI, despite the inherent competitive pressures from rivals like DeepSeek. This initiative not only addresses competitive dynamics but also fulfils a growing user demand for more understandable AI interactions, marking a significant stride towards more open artificial intelligence technology (¹).

OpenAI's approach to transparency involves revealing more of its AI's reasoning process while maintaining a coherent balance between safety and openness. By implementing comprehensive post‑processing mechanisms to filter harmful content, the o3‑mini model can present more accurate and safe outputs. Additionally, the model simplifies complex ideas and extends these transparency benefits across multi‑language barriers, broadening access to non‑English users (¹).

Despite these advancements, the transparency is not exhaustive; OpenAI refrains from divulging the full raw thought processes. This selective transparency, characterized by detailed summaries instead of unprocessed data, has sparked a dialogue about the extent to which AI models should reveal their decision‑making logic. This partial transparency is critiqued by stakeholders and experts who highlight the potential for 'blind spots' as a result of post‑processing filters (source).

The move towards enhanced transparency in AI technology has broader implications, potentially influencing regulatory environments and accelerating the development of new standards for AI governance. As companies embrace more transparent practices, they not only enhance user trust but also face increased developmental costs aimed at auditing and ensuring the explainability of AI outputs. Moreover, the competition to offer transparent AI solutions, as seen between OpenAI's and DeepSeek's models, will likely shape industry standards and drive further innovation (¹).

Public reactions to these advancements reflect a complex mix of appreciation and skepticism. While users commend the more accessible insight into AI reasoning, there remain concerns about performance issues and the accuracy of the AI's reasoning summaries. The decision to extend these enhancements across all user tiers, both free and paid, underscores an effort to democratize AI transparency but continues to stir debates around pricing and transparency ideologies compared to rivals such as DeepSeek, which opts for an open‑source approach (source).

Reasoning models are revolutionizing the landscape of decision‑making by offering enhanced transparency into artificial intelligence (AI) processes. These models analyze data and present outcomes with detailed breakdowns of their logical reasoning steps. For instance, the updated o3‑mini model by OpenAI, as noted in,¹ offers transparency by revealing more of its thought processes, which not only boosts user trust but also aligns with demands for greater interpretability from users and competitive models like DeepSeek's R1 model.

A vital aspect of these reasoning models is their capability to perform self‑fact‑checking before presenting results, thereby enhancing accuracy while minimizing errors. This is particularly significant as it helps differentiate between simple models and more advanced reasoning systems that invest in self‑validation processes. According to insights from OpenAI's o3‑mini updates, such self‑checking allows for more reliable outputs despite a trade‑off in response speed, which is essential for ethical AI applications.¹

By employing detailed post‑processing techniques to eliminate unsafe content, reasoning models like o3‑mini ensure a higher standard of safety and compliance. This method of post‑processing, while sparking debate about potential "blind spots," as discussed by Dr. Elena Rodriguez , is vital for maintaining ethical standards in AI deployment. The implementation of such measures is indicative of a broader industry trend towards responsible AI development, demonstrating a commitment to balancing transparency with proprietary protection and competitive market dynamics.

Despite these advancements, complete transparency remains elusive. OpenAI's decision to limit the dissemination of full raw data while providing enhanced summaries highlights the ongoing tension between maintaining intellectual property rights and meeting transparency expectations. As emphasized by Noam Brown, although the summaries do not encompass full thought detailing, the progress made is notable, aligning with efforts to address previous inaccuracies in AI outcomes.¹

Looking ahead, reasoning models are poised to significantly enhance decision‑making across various sectors. The incorporation of multi‑language support and the dissemination of transparent reasoning not only democratize access to advanced AI features but also prepare the foundation for future innovations in AI applications globally. As revealed by public responses, these updates in AI reasoning transparency, as seen with OpenAI's o3‑mini, are likely to increase trust and wider adoption of AI technologies.¹

Despite OpenAI's ongoing efforts to enhance the transparency of its AI models, significant limitations in current transparency measures remain. One major constraint is the selective disclosure of reasoning processes. OpenAI's recent update to the o3‑mini model aims to provide more detailed reasoning summaries, yet it still holds back from revealing the complete raw chain of thought due to competitive concerns and intellectual property protection. This selective transparency reflects a broader industry challenge where companies like DeepSeek exert pressure by offering more open and comprehensive models, forcing competitors to balance transparency with the need to protect proprietary information. More details can be found.¹

Transparency measures in AI also face technical hurdles. The introduction of post‑processing techniques to enhance safety by removing unsafe content can inadvertently lead to "blind spots," as cautioned by AI ethicist Dr. Elena Rodriguez. This filtering might obscure the very insights needed for full transparency, raising questions about the trade‑offs between safety and openness in AI systems. Such measures, while improving user trust, can create gaps that are not visible to end users, as further discussed here.

Another significant limitation is the accessibility and interpretation of the provided summaries. Although OpenAI has democratized these features for both free and paid tiers, users have mixed reactions regarding the clarity and reliability of these summaries. Some advanced users remain skeptical, demanding more robust verification mechanisms to ensure the AI's reasoning is accurate and comprehensive. This public sentiment emphasizes the necessity for transparency tools that not only inform but also instill user confidence effectively. For more insights, visit.¹

In the realm of AI development, transparency is not just about technology but also about policy and regulation. The balance between innovation protection and the growing demand for transparency is critical, especially in light of new international frameworks motivated by transparency leaders like OpenAI and DeepSeek. As industries adjust to these evolving standards, the competition between partial and full disclosure models will likely shape future regulatory landscapes. OpenAI's measured approach, for instance, could potentially set precedents in regulatory practices, as emphasized by legal experts discussed here.

User accessibility in AI models, like the OpenAI's o3‑mini, has seen significant advancements with the introduction of more detailed reasoning summaries. These detailed summaries aim to make AI's decision‑making processes more understandable for everyday users by providing insights into the underlying thought processes of the AI model. By doing so, OpenAI is enabling users to gain a deeper understanding of how and why specific outputs are generated, thereby enhancing transparency and trust [1](https://techcrunch.com/2025/02/06/openai‑now‑reveals‑more‑of‑its‑o3‑mini‑models‑thought‑process/).

Configuration options have been expanded to cater to both free and paid users, ensuring broad access to these transparency features. Paid users using the o3‑mini‑high configuration can access even more comprehensive summaries and advanced features, thus offering tailored experiences based on user needs and preferences. This flexibility in configuration not only democratizes access to advanced AI functionalities but also enhances user empowerment by allowing users to choose configurations that best meet their individual requirements [1](https://techcrunch.com/2025/02/06/openai‑now‑reveals‑more‑of‑its‑o3‑mini‑models‑thought‑process/).

In aligning the model with multi‑language support, OpenAI has improved its accessibility across non‑English speaking regions, thereby breaking down linguistic barriers and promoting a more inclusive user base. This adaptation is significant in that it allows users from diverse backgrounds to interact with AI models in their native languages, thus extending OpenAI's reach and utility globally. By incorporating translations into the updated model, OpenAI addresses the global demand for inclusivity and relevance in AI applications [1](https://techcrunch.com/2025/02/06/openai‑now‑reveals‑more‑of‑its‑o3‑mini‑models‑thought‑process/).

Post‑processing steps to remove unsafe content further demonstrate OpenAI's commitment to safety and user protection. These measures are crucial in creating a secure environment where users can interact with the AI without fear of encountering harmful or inappropriate content. Such safety protocols align with broader industry trends towards responsible AI development, ensuring that OpenAI's models not only perform powerfully but also responsibly [1](https://techcrunch.com/2025/02/06/openai‑now‑reveals‑more‑of‑its‑o3‑mini‑models‑thought‑process/).

In the rapidly evolving landscape of AI development, OpenAI's strategic decision to enhance transparency in its o3‑mini AI model has placed it on a competitive collision course with industry players such as DeepSeek. The recent updates to the o3‑mini model, which reveal more of its reasoning process, signal OpenAI's response to competitive pressures, notably from DeepSeek's R1 model. The R1 model is renowned for offering a full disclosure of its thought processes, making it a formidable rival in the transparency arena. OpenAI's updates aim to bridge this gap by providing detailed summaries of the reasoning steps taken by the AI, while also incorporating a multi‑language support system to widen accessibility. Such expansions are crucial for maintaining a competitive edge [1](https://techcrunch.com/2025/02/06/openai‑now‑reveals‑more‑of‑its‑o3‑mini‑models‑thought‑process/).

DeepSeek, one of the frontrunners pushing the boundaries in AI transparency, has recently launched its R2 model. This latest iteration goes beyond merely explaining AI reasoning by incorporating real‑time verification of decision‑making processes, thereby setting a new industry benchmark. While OpenAI has made strides towards transparency, it still stops short of full disclosure due to concerns over proprietary information. In contrast, DeepSeek's approach to transparency places it ahead in the race to win user trust and industry leadership. This competitive dynamic could potentially drive further innovation across the sector, as companies strive to balance transparency with intellectual property protection [1](https://www.deepseek.ai/blog/R2‑release).

While OpenAI and DeepSeek showcase differing approaches to transparency, other industry players are also making significant strides. For instance, Google DeepMind has introduced new safety protocols aimed at filtering content and detecting bias, addressing industry‑wide calls for responsible AI development. These moves underscore a broader industry trend towards enhanced AI transparency and accountability. Meanwhile, companies like Anthropic with their Claude 3.0 are advancing thought verification technologies that cross‑reference AI reasoning with internal databases to ensure accuracy. These diverse strategies reflect a burgeoning competitive landscape where transparency is pivotal [2](https://deepmind.google/safety‑standards‑2025) [3](https://www.anthropic.com/blog/claude‑3‑verification).

The competitive skirmishes within the AI sector concerning transparency are not merely about gaining immediate advantages but also about shaping the future of AI standards and practices. Initiatives like the EU AI Safety Coalition mandate for transparency documentation and Microsoft's integration of real‑time AI reasoning tracking tools are indicative of the growing regulatory and consumer demand for open AI systems. This push for transparency is creating a challenging environment for AI companies that must now innovate to meet new standards while fending off competition. However, this competitive pressure could lead to significant long‑term strides in AI accountability and safety, beneficial for both industry players and consumers alike [4](https://ec.europa.eu/ai‑safety‑standards) [5](https://azure.microsoft.com/blog/transparency‑tools).

OpenAI's recent updates to its o3‑mini model have caused quite a stir in the public domain, as both excitement and criticism have surfaced regarding the enhanced transparency features. Many users have applauded OpenAI for making AI processes more understandable through detailed reasoning summaries. The decision to extend these improvements to both free and paid tiers has been widely praised as a move towards democratizing AI access, highlighting OpenAI's commitment to inclusivity [4](https://techcrunch.com/2025/02/06/openai‑now‑reveals‑more‑of‑its‑o3‑mini‑models‑thought‑process/).

However, not all feedback has been positive. Critics on various tech forums have emphasized performance inconsistencies, especially during peak usage times. There is an apparent preference among some users for DeepSeek's R1 model, which offers a fully open‑source approach despite potential security risks. This debate between transparency and security continues to echo throughout the tech community, with many raising concerns about how these updates might affect user trust and pricing dynamics [6](https://opentools.ai/news/openai‑unveils‑transparent‑reasoning‑in‑o3‑mini‑amidst‑deepseek‑competition).

Internationally, OpenAI's introduction of multilingual support and improved safety features via a post‑processing step has been well‑received. These updates have won particular acclaim from international users and safety advocates, who appreciate the broader accessibility and enhanced content safety. Nevertheless, a segment of technical users remains skeptical, positing that the reasoning summaries' accuracy needs further verification mechanisms to ensure reliability and trust [8](https://www.datagrom.com/ai‑news/openai‑enhances‑o3‑minis‑thought‑transparency.html).

These mixed perceptions indicate that while OpenAI's transparency enhancements represent progress, they also bring to light ongoing issues that need to be addressed. Continued public dialogue and critique will likely shape future updates and influence the direction of AI development. As AI systems become more transparent, balancing user privacy, intellectual property rights, and competitive strategy will remain a challenge of paramount importance for OpenAI [5](https://techcrunch.com/2025/02/06/openai‑now‑reveals‑more‑of‑its‑o3‑mini‑models‑thought‑process/).

The increased transparency in AI models, as evidenced by OpenAI's recent updates, is set to significantly influence AI regulations and standards. By providing detailed reasoning summaries of the o3‑mini model, OpenAI is responding to both competitive pressures from companies like DeepSeek and the growing demand for transparency in AI operations. These changes align with global efforts to establish more robust AI governance, pushing for a balance between innovation and ethical responsibility [TechCrunch link](https://techcrunch.com/2025/02/06/openai‑now‑reveals‑more‑of‑its‑o3‑mini‑models‑thought‑process/). This move may inspire other companies to follow suit, incorporating transparency into their AI systems as a fundamental feature.

As the AI industry progresses towards greater transparency, regulatory bodies and international coalitions might expedite the formulation of new standards. For instance, the EU AI Safety Coalition's initiatives mandating documentation of AI decision‑making processes signify a trend towards heightened accountability in AI development [EC link](https://ec.europa.eu/ai‑safety‑standards). Such regulations are likely to influence global market dynamics, as companies must adapt their technologies and processes to remain compliant. These measures could lead to increased consumer trust, enabling more widespread adoption of AI technologies across sectors.

Moreover, transparency in AI does not only benefit regulatory compliance but also supports the industry’s economic and social transformation. With OpenAI's adjustments, AI models like the o3‑mini can empower users by providing insight into decision‑making processes, thereby fostering trust and reducing cognitive bias [OpenTools link](https://opentools.ai/news/openais‑o3‑mini‑boosts‑transparency‑with‑new‑chain‑of‑thought‑reveal). However, this newfound openness might also expose systems to potential adversarial attacks, requiring companies to invest in more sophisticated security protocols.

The ongoing competition between AI firms, especially regarding transparency features, will likely set new precedents. While OpenAI opts for partial disclosure to protect intellectual property, competitors like DeepSeek favor full transparency, creating differing industry benchmarks. This divergence highlights the tension between maintaining competitive advantage and meeting public and regulatory demands for openness [DataGrom link](https://www.datagrom.com/ai‑news/openai‑enhances‑o3‑minis‑thought‑transparency.html). The result may be a more nuanced and collaborative approach to AI development, with organizations aligning their tactics with emerging legal and ethical standards.

As the landscape of artificial intelligence continues to evolve, transparency has emerged as a cornerstone for ensuring that AI systems are trustworthy and reliable. OpenAI's recent updates to its o3‑mini models, detailed in a,¹ demonstrate the company's commitment to transparency. This move entails revealing more about the model's reasoning processes, which is expected to enhance trust among users while also pushing competitors to follow suit. Such transparency is not just about user reassurance; it's pivotal for accelerating the economic adoption of AI, as businesses and industries are more likely to integrate systems they understand.

Economically, the implications of AI transparency are profound. By clarifying AI decision‑making processes, companies can drive efficiency and productivity gains, though these benefits come with increased development costs. Implementing transparency measures requires substantial investment in explainability and auditing tools. According to analysts, this could lead to market consolidation where only larger companies with the necessary resources can keep pace with evolving transparency requirements. On a broader scale, transparency can foster inclusivity by democratizing access to AI capabilities, thus reshaping market dynamics.

On the social front, transparency in AI can empower users by offering greater insights into AI's decision‑making, thereby reducing biases and discriminatory practices inherent in opacity . However, with greater transparency comes the risk of exposing systems to adversarial attacks, as more details about system internals become available. Ensuring that transparency does not compromise security is crucial. The enhanced understanding that users gain through transparency not only boosts trust but also fuels more informed interactions with technology, aligning with broader societal calls for ethical AI practices.

Politically, the push for AI transparency has triggered a reassessment of regulations and standards governing AI technologies. As international frameworks for AI governance evolve, the tension between protecting intellectual property and fulfilling transparency expectations continues to grow. In this regulatory landscape, the EU AI Safety Coalition's initiatives are crucial, as they mandate detailed documentation of AI models' decision processes, setting a precedent for accountability . This regulatory momentum is expected to prompt more companies to align their practices with global standards, ensuring that transparency efforts translate into real‑world accountability.

The competitive dynamics between key players like OpenAI, DeepSeek, and others, as they experiment with different levels of transparency, are likely to shape the future of AI standards. OpenAI's approach of partial disclosure contrasts with DeepSeek’s full model transparency, as reflected in their R2 model release . Such competition is essential for driving innovation in responsible AI development. Observing how these transparency paradigms influence public perception and regulatory requirements will provide valuable insights into how best to integrate transparency in AI without hampering innovation or security.

OpenAI's journey towards greater transparency, as exemplified by the o3‑mini model update, marks a significant milestone in the AI landscape. By revealing more of the model's reasoning process, OpenAI addresses longstanding demands from users and industry rivals alike for clearer insights into AI decision‑making. Although the company has not unveiled every nuance of its algorithm, the enhanced summaries signify a step in the right direction towards fostering user trust. This strategic move, primarily driven by competition, also highlights the broader industry shift towards accountability. As OpenAI fine‑tunes the balance between transparency and maintaining its proprietary edge, one can't help but wonder how this will influence future AI development across the board. Indeed, the increased transparency could set new expectations for AI ethics and operational standards, influencing industry practices significantly going forward [1](https://techcrunch.com/2025/02/06/openai‑now‑reveals‑more‑of‑its‑o3‑mini‑models‑thought‑process/).

While the advancements in transparency afford users a clearer view of AI models' "thought processes", they also pose complex challenges. Enhanced reasoning summaries and multilingual capabilities offer considerable benefits, yet, they come bundled with concerns over performance reliability and verified accuracy. Public reaction varies, with some praising the move as overdue progress, while others remain skeptical, citing competitors like DeepSeek with their open‑source approach as more reliable in terms of transparency. Furthermore, as new features roll out, such as post‑processing filters for content safety, the discussion veers towards ethical considerations around censorship and potential bias, igniting debate about the true essence of "transparent AI". As OpenAI paves this road, the tech community and policymakers alike will be keenly observing how these updates impact public perception and regulatory standards [1](https://techcrunch.com/2025/02/06/openai‑now‑reveals‑more‑of‑its‑o3‑mini‑models‑thought‑process/).

In conclusion, OpenAI's journey towards augmented transparency is a crucial reflection of the evolving relationship between AI developers, users, and society at large. It's a testament to the increasing demands for ethical AI that not only delivers accurate and safe content but also holds itself accountable to its audience. This balance between innovation and transparency invites further speculation about the competitive dynamics within AI industries, especially when juxtaposed against organizations like DeepSeek, which favor complete openness. The lessons learned and adjustments made by OpenAI will undeniably influence the future of AI developments, underlining the importance of transparency in fostering not just trust, but responsible AI ecosystems that can adapt to emerging ethical standards and continue to serve diverse global audiences effectively [1](https://techcrunch.com/2025/02/06/openai‑now‑reveals‑more‑of‑its‑o3‑mini‑models‑thought‑process/).