Manus AI and DeepSeek: How Do These Chinese AIs Stack Up Against Grok 3 and ChatGPT

Manus AI and DeepSeek: How Do These Chinese AIs Stack Up Against Grok 3 and ChatGPT- Mar 13, 2025- 6 min readLet do a comparison of Manus AI, Grok 3, DeepSeek R1, and ChatGPT (including o3-mini and GPT-4o), based on their capabilities. Each model was evaluated across six key categories: Reasoning and Problem-Solving, Real-Time Data Access, Coding and Execution, Versatility and Creativity, Accessibility and Cost, and Speed. The analysis draws from recent benchmarks, public documentation, and industry reports, ensuring a thorough understanding for both technical and non-technical audiences.

Background and Context- Manus AI, launched on March 6, 2025, by the Chinese startup Monica, is a fully autonomous AI agent designed to execute real-world tasks end-to-end, such as travel planning and stock analysis (What is Manus? China's World-First Fully Autonomous AI Agent Explained). It has gained attention for its performance on the GAIA benchmark, with scores of 86.5% (Level 1), 70.1% (Level 2), and 57.7% (Level 3) (Manus AI Statistics and Facts).

• Grok 3, released by xAI in February 2025, is a reasoning-focused model with advanced real-time data access via DeepSearch, scoring 93.3% on the AIME math benchmark (Grok 3 Beta — The Age of Reasoning Agents | xAI). It is tied to X Premium+ ($40/month) or rumored SuperGrok ($30/month) plans.

• DeepSeek R1, from DeepSeek AI, is an open-source reasoning model launched in January 2025, known for its efficiency and cost-effectiveness, with a free tier and scores like 71.0% on AIME 2024 (DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning).

• ChatGPT, developed by OpenAI, includes o3-mini (a cost-efficient reasoning model with 87.3% on AIME high version) and GPT-4o (a versatile multimodal model), with access ranging from free to $200/month for Pro plans (OpenAI o3-mini: Performance, How to Access, and More).

Image from The Register#### Category-by-Category Analysis### Reasoning and Problem-SolvingThis category evaluates models on their ability to handle complex reasoning tasks, primarily using the AIME math benchmark for consistency, with GAIA as a secondary measure for real-world problem-solving.

• Grok 3: Achieves 93.3% on AIME, indicating strong mathematical reasoning, and is designed for versatile problem-solving (Grok 3 Beta — The Age of Reasoning Agents | xAI).

• Manus AI: No specific AIME score, but excels on GAIA with 86.5% (Level 1), 70.1% (Level 2), and 57.7% (Level 3), suggesting robust real-world reasoning (Manus AI Statistics and Facts).

• DeepSeek R1: Scores 71.0% Pass@1 on AIME 2024, showing solid technical reasoning but lagging behind top models (DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning).

• ChatGPT o3-mini: High version scores 87.3% on AIME, competitive for reasoning tasks, with a focus on STEM (OpenAI o3-mini: Performance, How to Access, and More).

Winner: Grok 3, due to its highest AIME score, reflecting superior reasoning capabilities.### Real-Time Data AccessThis category assesses the models' ability to fetch and integrate current information, crucial for dynamic tasks.

• Grok 3: Features DeepSearch mode for real-time web and X searches, pulling fresh info instantly, enhancing its responsiveness (Elon Musk’s xAI releases its latest flagship model, Grok 3 | TechCrunch).

• Manus AI: Likely has real-time data access for real-world tasks, given its autonomous execution capabilities, though specifics are unclear (Manus AI: Capabilities, GAIA Benchmark Insights, Use Cases & More).

• DeepSeek R1: Offers web browsing, but reports suggest it struggles under high demand, limiting its real-time effectiveness (DeepSeek - R1 Online (Free|Nologin)).

• ChatGPT o3-mini: Includes search integration for real-time data, with early prototype support, enhancing its utility (OpenAI O3-Mini: The Cost-Efficient Genius Redefining STEM AI | Medium).

Winner: Grok 3, with its advanced DeepSearch mode providing the most integrated real-time data access.### Coding and ExecutionThis category evaluates coding proficiency and the ability to execute tasks autonomously, using benchmarks like LiveCodeBench where available.

• Manus AI: Excels in autonomous execution, building functional outputs like websites and games, with no specific benchmark scores but strong real-world performance (China’s Autonomous Agent, Manus, Changes Everything | Forbes).

• Grok 3: Scores 79.4% on LiveCodeBench, beating GPT-4o (72.9%), indicating strong coding capabilities (Grok 3 Beta — The Age of Reasoning Agents | xAI).

• DeepSeek R1: Achieves 57.2% on LiveCodeBench, with distilled models performing well in coding tasks, but overall lower than top models (DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning).

• ChatGPT o3-mini: High version scores 0.846 on LiveBench coding average, suggesting strong coding performance, though benchmark specifics vary (o3-mini Early Days — LessWrong).

Winner: Manus AI, due to its superior execution capabilities, surpassing others in practical task completion.### Versatility and CreativityThis category assesses the models' ability to handle diverse tasks, including creative writing and open-ended chats, considering ChatGPT's GPT-4o for its multimodal strengths.

• Grok 3: Handles technical tasks and creative writing, with a focus on humor and open-ended chats, making it versatile (Elon Musk debuts Grok 3, an AI model that he says outperforms ChatGPT and DeepSeek | CNN Business).

• ChatGPT (GPT-4o): Highly versatile and creative, excelling in multimodal tasks like image and text generation, with polished prose (GPT-4o vs GPT-4o Mini: Choosing the Right AI Model | Amity Solutions).

• Manus AI: Focused on practical execution, less on creativity, with limited chat capabilities (Manus AI: Features, Architecture, Access, Early Issues & More | DataCamp).

• DeepSeek R1: Weak on creativity, primarily technical, with dry responses (DeepSeek R1 Review: Performance in Benchmarks & Evals | TextCortex).

Winner: Tie between Grok 3 and ChatGPT (GPT-4o), both excelling in versatility and creativity, with GPT-4o slightly ahead in multimodal tasks.### Accessibility and CostThis category evaluates ease of access and pricing, crucial for user adoption.

• DeepSeek R1: Offers a free tier, open-source weights under MIT license, making it highly accessible, with API pricing at $0.14/million input tokens (cache hit) (DeepSeek R1 is now available on Azure AI Foundry and GitHub | Microsoft Azure Blog).

• ChatGPT: Free base model, Plus plan at $20/month for o3-mini access, Pro at $200/month, offering broad accessibility (Announcing the availability of the o3-mini reasoning model in Microsoft Azure OpenAI Service | Microsoft Azure Blog).

• Grok 3: Tied to X Premium+ at $40/month or rumored SuperGrok at $30/month, limited to X ecosystem, less accessible (Grok 3 AI is now free to all X users – here's how it works | ZDNET).

• Manus AI: Invite-only, with codes reselling for up to $7,000 USD, likely premium, least accessible (Manus AI Statistics and Facts).

Winner: DeepSeek R1, due to its free tier and open-source nature, offering the best cost-effectiveness.### SpeedThis category measures response and processing speed, vital for user experience.

• Grok 3: Described as lightning-fast, with scripts and searches in seconds, leveraging xAI’s 100,000+ GPU backbone (Elon Musk’s ‘Scary Smart’ Grok 3 Release—What You Need To Know | Forbes).

• Manus AI: Demos suggest fast for complex tasks, but no specific metrics (Another DeepSeek moment? General AI agent Manus shows ability to handle complex tasks | South China Morning Post).

• DeepSeek R1: Achieves 381 tokens/sec, outpacing many rivals, but web browsing may lag (DeepSeek - R1 Online (Free|Nologin)).

• ChatGPT o3-mini: Faster than o1-mini, with 2.5s faster time to first token, lower latency (OpenAI launches o3-mini, its latest 'reasoning' model | TechCrunch).

Winner: Grok 3, highlighted for its exceptional speed across tasks.Model Comparison by Artificial Analysis#### Overall AssessmentGrok 3 emerges as the most well-rounded model, winning in Reasoning and Problem-Solving, Real-Time Data Access, and Speed, with a tie in Versatility and Creativity alongside ChatGPT (GPT-4o). Manus AI excels in Coding and Execution, particularly for autonomous task completion, but its invite-only status limits accessibility. DeepSeek R1 offers the best Accessibility and Cost, appealing to budget-conscious users with its open-source nature. ChatGPT, through o3-mini and GPT-4o, provides a balanced suite, with GPT-4o standing out for creativity and versatility. The choice depends on specific user needs, with Manus AI's rapid market impact (invite codes reselling for up to $7,000 USD) highlighting its high demand despite limited access (Manus AI Statistics and Facts).

This analysis ensures a comprehensive understanding, drawing from benchmarks like AIME (Comparison of AI Models across Intelligence, Performance, Price | Artificial Analysis), GAIA (GAIA: a benchmark for General AI Assistants | arXiv), and LiveCodeBench (LiveCodeBench: Holistic and Contamination Free Evaluation of Large Language Models for Code | arXiv), among others, to provide a detailed comparison.

Tags:

• Grok3- AI- DeepSeek- ChatGPT- Manus- Tech & AI