What is good about DeepSeek

For users from the Russian Federation, It has an undeniable advantage over ChatGPT: free access. Chinese developers do not restrict requests from Russian IP addresses in any way. Even the mobile application can be downloaded without exercises with a change of region. The neural network understands requests in Russian and easily answers in it.

DeepSeek also features a step-by-step reasoning function, which is useful for solving logical and mathematical problems. In this regard, the Chinese model is similar to OpenAI’s o1 and o1 Pro. With one exception: Deep Think (that’s the name of the analytical mode in it) is available for free. The issue with the obsolescence of training data has also been resolved: the chatbot can go online.

Artificial intelligence has seen rapid advancements in recent years, and the latest contender making waves is DeepSeek V3, a Chinese AI model some are hailing as a “ChatGPT killer.” But is it truly a game-changer, or are there significant caveats that should temper our enthusiasm? After testing it extensively, I’ve found both impressive strengths and notable weaknesses. Here’s my full breakdown.

The DeepSeek team apparently plans to pay for their project through a paid API for developers. At the same time, the tariffs are many times more profitable than, for example, GPT-4o.

Its website has a table comparing their model with popular solutions from OpenAI, Anthropic, and other industry leaders. In many benchmarks, DeepSeek-V3 is either ahead of or almost as good as GPT-4o.

Why DeepSeek V3 is Garnering Attention

From the moment I logged into DeepSeek V3, I was struck by its minimalistic interface. It mirrors ChatGPT in many ways: you can track your conversation history, start new chats, and toggle between light and dark modes. However, a few standouts feature immediately set it apart:

1. Large File Handling: DeepSeek can process up to 50 documents with a maximum file size of 100 MB each, dwarfing the capabilities of ChatGPT or Claude.
2. Enhanced Reasoning with DeepSync: Leveraging a “chain-of-thought” process, DeepSeek tackles complex problems by breaking them into smaller steps. This is particularly useful for tasks involving programming, mathematics, or practical daily applications.

When I first tested it, the response speed and quality were jaw-dropping. For example, I uploaded a basic PDF job description and asked for a summary. Not only was it lightning-fast, but the summary was clear and accurate. DeepSeek even suggested ways to improve the document — structuring it better, adding measurable KPIs, and specifying required skills.

Key Strengths

1. Performance in Multilingual Tasks

DeepSeek V3 excels in Chinese, but it also delivers strong results in multilingual tasks, addressing one of the pain points of many AI models that struggle outside English.

2. Efficient Architecture

Its “Mixture of Experts” (MoE) architecture is revolutionary. Unlike traditional models that activate all parameters for a task, it uses only the most relevant ones, reducing energy consumption while maintaining quality. With 671 billion parameters, it rivals leading models like GPT-4 but operates more efficiently.

3. Affordability

Its API pricing is incredibly competitive:

• $0.14 per million tokens (input)
• $0.28 per million tokens (output)

For comparison, Amazon’s Nova Pro AI charges $3.20 per million tokens for output, making it more than 10 times cheaper.

But Is It Better than ChatGPT?

While DeepSeek is impressive, calling it a “ChatGPT killer” is an overstatement. I came across a comparison of its performance to ChatGPT-4.0 Pro using complex queries. One test involved identifying a protein meeting specific criteria. DeepSeek struggled to deliver accurate results, unlike ChatGPT, which nailed the answer with a detailed explanation.

Moreover, when tackling personal development questions — like strategies for avoiding burnout — DeepSeek’s responses felt generic, lacking the depth ChatGPT offers, especially when personalized through its memory feature.

Red Flags: Security and Data Privacy

Its efficiency comes at a potential cost: data privacy concerns. Here are a few critical points:

1. Chinese Jurisdiction: User data is governed by the laws of the People’s Republic of China. For European and American users, this means limited recourse in case of disputes or data misuse.
2. Lack of Transparency: Unlike GDPR-compliant models, DeepSeek offers minimal information about how user data is stored and used.
3. Unlimited Data Retention: Conversations, uploaded files, and other data are retained indefinitely — something users should be cautious about.

Recommendation: If you decide to test DeepSeek, use a throwaway email and unique password. Avoid sharing sensitive or personal data.