DeepSeek-R1

DeepSeek-R1 is a family of open-source large language models (LLMs) developed by the Chinese artificial intelligence company DeepSeek. Released in January 2025, the models are specifically designed to excel at complex reasoning tasks, including mathematics, coding, and logical deduction. The flagship model, DeepSeek-R1, contains 671 billion total parameters with 37 billion active parameters per token, employing a Mixture-of-Experts (MoE) architecture that activates only a subset of its parameters for each input. The model was released alongside a technical report and made available under the MIT License, allowing free distillation and commercialization.

The release of DeepSeek-R1 marked a significant moment in the development of AI reasoning models. It was the first open research to demonstrate that advanced reasoning capabilities in LLMs could be incentivized purely through reinforcement learning (RL), without relying on supervised fine-tuning (SFT) on human-annotated reasoning trajectories. This approach contrasted sharply with prevailing methods that depended heavily on extensive human-labelled data to teach models how to reason step by step.

Architecture

DeepSeek-R1 is built upon the DeepSeek-V3 Base model and utilizes a Mixture-of-Experts (MoE) framework. The architecture incorporates Multi-head Latent Attention (MLA) and a specialized DeepSeekMoE structure. It also includes built-in multi-token prediction (MTP) for speculative decoding, which improves inference efficiency. The model supports a context length of 128,000 tokens, enabling it to process and reason over long documents and complex problem statements.

The MoE design allows DeepSeek-R1 to maintain a large total parameter count while keeping computational costs manageable during inference. Only 37 billion parameters are activated for any given token, which is a fraction of the total 671 billion. This architecture enables the model to deliver high performance on reasoning-intensive tasks without the prohibitive inference costs typically associated with dense models of comparable scale.

Training Methodology

The training pipeline for DeepSeek-R1 represents a departure from conventional LLM development. The researchers first created DeepSeek-R1-Zero, a model trained via large-scale reinforcement learning directly on the base model, bypassing the supervised fine-tuning stage entirely. The reward signal was based solely on the correctness of final answers, without imposing any constraints on the reasoning process itself.

Through this pure RL training, DeepSeek-R1-Zero spontaneously developed sophisticated reasoning behaviors, including self-verification, reflection, and the exploration of alternative solution paths. The model exhibited a tendency to generate longer responses that incorporated verification steps and backtracking when encountering dead ends. This emergent behavior demonstrated that reasoning capabilities could arise from RL without explicit human instruction on how to reason.

However, DeepSeek-R1-Zero suffered from several practical issues, including poor readability, endless repetition, and language mixing, where the model would occasionally switch between English and Chinese within a single chain of thought. To address these problems, the researchers introduced DeepSeek-R1, which incorporates a small amount of cold-start data before RL training. The final pipeline involves two RL stages aimed at discovering improved reasoning patterns and aligning with human preferences, along with two SFT stages that serve as seeds for the model's reasoning and non-reasoning capabilities.

Performance and Benchmarks

DeepSeek-R1 achieves performance comparable to OpenAI's o1 model across a range of reasoning benchmarks. On the AIME 2024 mathematics competition, DeepSeek-R1 attained a pass@1 score of 79.8 percent, while OpenAI's o1-1217 scored 79.2 percent. On MATH-500, the model achieved 97.3 percent pass@1, slightly surpassing o1's 96.4 percent. For coding tasks, DeepSeek-R1 reached a Codeforces rating of 2029, outperforming o1's 1891 rating. On the GPQA Diamond benchmark, which tests graduate-level scientific reasoning, DeepSeek-R1 scored 71.5 percent pass@1 compared to o1's 75.7 percent.

Beyond the flagship model, DeepSeek released six distilled dense models based on the Qwen and Llama architectures. These smaller models, ranging from 1.5 billion to 70 billion parameters, were fine-tuned using reasoning data generated by DeepSeek-R1. The DeepSeek-R1-Distill-Qwen-32B model outperformed OpenAI's o1-mini across various benchmarks, establishing new state-of-the-art results for dense models. The 70-billion-parameter distilled model achieved an AIME 2024 pass@1 score of 70.0 percent, demonstrating that reasoning patterns from large models can be effectively transferred to smaller ones.

Open-Source Release and Distillation

DeepSeek-R1 was released under the MIT License, making it one of the most permissively licensed frontier reasoning models available. The open-source release included the full model weights for both DeepSeek-R1 and DeepSeek-R1-Zero, along with the six distilled models. The technical report was published on arXiv and later in the journal Nature, providing the research community with detailed insights into the training methodology and architecture.

The decision to open-source the model and its distilled variants had significant implications for the AI research community. Researchers could now study and build upon a reasoning model that rivaled proprietary systems, and developers could fine-tune the models for specialized applications. The MIT License explicitly permitted distillation and commercialization, removing legal barriers that often accompany the release of advanced AI models.

Impact and Reception

The release of DeepSeek-R1 generated considerable attention in both the AI research community and the broader technology industry. Its performance on par with OpenAI's o1 model, combined with its open-source nature, challenged assumptions about the necessity of massive proprietary investment for developing frontier reasoning capabilities. The model demonstrated that pure reinforcement learning could incentivize sophisticated reasoning behaviors without the need for extensive human-labelled reasoning data, a finding that has since influenced subsequent research in the field.

The distilled models released alongside DeepSeek-R1 also had a notable impact, particularly for researchers and developers with limited computational resources. The DeepSeek-R1-Distill-Qwen-32B model, for instance, achieved performance exceeding OpenAI's o1-mini on several benchmarks despite being substantially smaller and more efficient to run. This made advanced reasoning capabilities accessible to a wider range of users and applications.

The model's architecture and training methodology have been widely studied and replicated. The use of Group Relative Policy Optimization (GRPO) as the RL framework, which eliminates the need for a separate value network, has been adopted in other reasoning model projects. The concept of using pure RL to incentivize reasoning, bypassing SFT on human demonstrations, has become a prominent research direction.

References

DeepSeek API Docs. "DeepSeek-R1 Release." 

Nature. "DeepSeek-R1 incentivizes reasoning in LLMs through reinforcement learning." 

SGLang Documentation. "DeepSeek-R1." 

Hugging Face. "README.md · qgallouedec/DeepSeek-R1 at main." 

[Read more →]

ChatGPT

ChatGPT is an artificial intelligence (AI) model developed by OpenAI, designed to generate text similar to human writing based on input provided by the user. Launched in November 2022, ChatGPT quickly became a global phenomenon and changed the way society interacts with technology, from professional work to everyday personal use. Its presence has not only revolutionized productivity but also sparked in-depth discussion about the future of knowledge, education, and social interaction in the digital era.

History and Development

The technology underlying ChatGPT is OpenAI's GPT (Generative Pre-trained Transformer) model series, which has undergone a number of significant improvements. Since its debut with GPT-3.5, the model has transformed through various versions up to GPT-5.5, which has become the most frequently used model. Each new generation has brought improvements in reasoning ability, following of complex instructions, and overall conversation quality.

This development marks a shift from merely a language model to a more intelligent assistant. In mid-2026, models such as GPT-4.5 and GPT-5.2 were officially retired in ChatGPT and shifted to newer versions to ensure users receive the best experience. In addition, OpenAI periodically releases new models such as GPT-5.5 Instant Mini, which functions as a backup model when users reach usage rate limits, ensuring continued access to the service.

Main Features and Capabilities

ChatGPT has evolved from merely a chat box into an advanced platform with a variety of features designed to increase productivity and personalization.

Context and Personalization

One of ChatGPT's main capabilities is memory and personalization. This feature allows ChatGPT to remember information from previous conversations, providing more relevant and contextual responses over time. Users can view a memory summary, correct stored information, or turn off the memory function entirely to maintain privacy. This personalization makes ChatGPT an assistant that is increasingly adaptive to each user's specific needs.

Scheduled Tasks

The "Scheduled Tasks" feature allows users to ask ChatGPT to perform repeating work, send reminders, or monitor something at a predetermined time. This changes ChatGPT from a reactive tool into a proactive tool that can be relied on for various daily routine needs.

Codex and Work Automation

The development of Codex marked a major leap in ChatGPT's operational capabilities. Features such as Codex Remote allow users to control a remote computer from a phone, while Record & Replay allows users to demonstrate a workflow and turn it into a reusable skill. For business users, this opens large opportunities to automate repeating tasks in various fields, such as software development, data analysis, and project management. The Developer mode for Browser use feature also gives developers more control for debugging and performance analysis.

Productivity and Accessibility Improvements

In addition to the features above, ChatGPT has consistently introduced improvements to the user experience. Improvements to the dictation model with a word error rate lower by up to 10% make it easier for users to interact without typing. For users who enter long text, ChatGPT now automatically converts pastes of more than 10,000 characters into attachments, keeping the chat box tidy. Pronunciation assistance features for more than 60 languages and the integration of current information updates such as World Cup schedules also add to ChatGPT's usefulness.

Impact on Knowledge and Education

The presence of ChatGPT has had a major impact on the way humans access and process information. As a language model trained on large amounts of data from the internet, including Wikipedia articles, ChatGPT is able to provide instant and concise answers to user questions. This has changed information-seeking behavior, in which users can obtain knowledge without having to visit the source site directly, such as Wikipedia, which is one of its main training data sources. A study even showed that Wikipedia is the source most frequently cited by ChatGPT, with a share reaching 16.3% of all citations.

Role in Education

In the world of education, empirical studies have explored ChatGPT's potential as a cognitive scaffold. Research in 2026 at a madrasah in Indonesia showed that the use of ChatGPT in argumentative writing instruction had a large effect on student achievement, particularly in the aspects of organization and written content. ChatGPT was shown to help students construct a framework of thinking and develop arguments, although its effectiveness was not even across all linguistic aspects such as vocabulary and grammar. Another study also looked at the use of ChatGPT in ecology-based Indonesian language learning, opening opportunities for more innovative teaching methods.

Knowledge Dynamics in the AI Era

Experts see that generative AI, including ChatGPT, will change the knowledge landscape. There are concerns that if AI becomes increasingly advanced, the role of encyclopedias such as Wikipedia as a center of knowledge could be displaced. However, there is also a view that AI can instead become a tool for democratizing knowledge, by translating and disseminating information into various regional languages that were previously difficult to access. This discussion highlights the importance of the role of human curators in ensuring the quality and validity of information, even if that content is generated by AI.

Regulation and Governance

Along with its growing popularity, ChatGPT and other digital platforms have faced oversight from regulators in various countries. In Indonesia, in November 2025, the Ministry of Communication and Digital Affairs (Komdigi) issued a warning to dozens of major digital platforms, including OpenAI (ChatGPT) and Wikipedia, because they had not yet registered as Private-Scope Electronic System Operators (PSE). This step was taken to ensure compliance with Minister of Communication and Informatics Regulation Number 5 of 2020, with the threat of sanctions in the form of access blocking if not fulfilled immediately.

Social Phenomenon: Confiding in AI

The phenomenon of ChatGPT use is not limited to a productivity tool, but has also entered the social-emotional realm. Digital sociology research in late 2025 revealed that many users in Indonesia use ChatGPT as a "confidant." The main factors behind this phenomenon are the perception that AI is non-judgmental, always available 24 hours a day, provides emotional validation, and offers a sense of safety in terms of privacy. However, this research also warned of potential negative impacts, such as emotional dependence, widening social distance, and the risk of data leaks. This highlights the need for balanced digital literacy so that AI does not replace social relations among humans.

References

· Wikimedia. (2025, Desember 15). Education/News/December 2025/WikipediaxAI: Wikipedia, AI, and the future of knowledge. Meta-Wiki. 

· OpenAI Help Center. (2026, Juni 26). ChatGPT — 版本说明. 

· OpenAI Help Center. (2026, Juli 6). ChatGPT — Release Notes. 

· OpenAI Help Center. (2026, Juni 26). ChatGPT Business - 发行说明. 

· OpenAI Help Center. (2026, Juni 26). ChatGPT Business - 發布說明. 

· Hops.ID. (2025, November 19). Cek HP Anda Sekarang! 25 Aplikasi Favorit Ini Terancam Hilang, Ada ChatGPT hingga Wikipedia. 

· Bennylin. (2023, Mei). Pengguna:Bennylin/20TahunWikipedia. Wikipedia bahasa Indonesia. 

· Syafawani, R., dkk. (2025, Desember 23). Fenomena Curhat ke AI: Tinjauan Sosiologi Digital Terhadap Pergeseran Interaksi Sosial di Indonesia. Jurnal Sosial dan Sains. 

· Jumaddil, J., dkk. (2026, April 30). ChatGPT as a Macro-Level Cognitive Scaffold for First-Language Argumentative Writing. Edcomtech. 

· Steavenson, M. (2025, Juni 14). AI Search Visibility: Why Reddit, Wikipedia, and YouTube Win And What It Means For Your Brand. LinkedIn. 

· Abdian Amien, A., & Kusumawati, H. (2024, Desember). Optimalisasi CHATGPT Dalam Pembelajaran Bahasa Indonesia Berbasis Ekologi Di SMAN 4 Pamekasan. GHANCARAN. 

[Read more →]