NEW DeepSeek V3 0324: DESTROYS Claude 3.7? 🤯

Key Concepts

Deepseek V3, Claude 3.7 Sonnet, Reasoning Model, Base Model, SEO Optimization, AI Detection, Humanization, Web Search, Open Source Model, API Key, Coding, Landing Page, User Intent, AI Agents, Automations.

Writing Ability

• Prompt: Create an SEO optimized article for "SEO training in Manchester."
• Deepseek V3: Frontloaded the keyword, humanized content, 485 words, 0% AI detectable.
• Claude 3.7 Sonnet: Included "2023" (incorrect), used AI fluff words ("top-notch"), 1,087 words, 0% AI detectable.
• Zero GPT AI Detection: Both models scored 0% AI detection, indicating highly humanized content.
• Conclusion: Deepseek V3 produced more natural and humanized content compared to Claude 3.7 Sonnet.

Reasoning Challenge

• Prompt: "There's a tree on the other side of a river in winter. How can I pick it up?"
• Claude 3.7 Sonnet: Faster response, recognized apples don't bear fruit in winter, but lacked in-depth analysis.
• Deepseek V3: Slower initial loading, but provided a more in-depth analysis of the situation, considered apple availability in winter, and offered multiple solutions.
• Conclusion: Deepseek V3 won due to its comprehensive analysis and logical reasoning.

Web Search

• Prompt: "What happened in AI news today?"
• Deepseek V3: Search service temporarily unavailable due to technical issues.
• Claude 3.7 Sonnet: Only available in the US.
• Open Router: Web search failed and still showed old date (June 2024).
• Conclusion: Both models failed the web search test. Claude was considered slightly better for being upfront about unavailability.

Open Source and API Key

• Deepseek V3: Open source model, can be hosted locally, free API key available on openrouter.ai.
• Claude 3.7 Sonnet: Requires payment for tokens, which can be expensive.
• Conclusion: Deepseek V3 offers a significant advantage with its free API key, enabling users to build AI agents and automations without cost.

Coding Challenge 1: Landing Page

• Prompt: "Create a beautiful landing page for my SEO agency."
• Deepseek V3: Created a well-designed, complete landing page with a footer, nice title headings, and icons.
• Claude 3.7 Sonnet: Created a thin page with a title bar, image, and blank space, lacking a footer and proper design.
• Conclusion: Deepseek V3 outperformed Claude 3.7 Sonnet in creating a visually appealing and functional landing page.

Coding Challenge 2: Sleek Landing Page

• Prompt: "Create a sleek conversion-focused landing page for Goldie agency's automation dashboard. The design should be modern etc."
• Deepseek V3: Failed to recognize the coding task.
• Claude 3.7 Sonnet: Correctly identified the task as coding and started generating code.
• Conclusion: Claude 3.7 Sonnet demonstrated better user intent recognition in this coding task.

Overall Conclusion

Deepseek V3 is a strong contender, especially considering its free availability and open-source nature. It excels in content creation, reasoning, and provides a free API key. However, Claude 3.7 Sonnet offers superior UI, higher quality content, and a more intelligent model overall. The choice depends on the user's priorities: cost-effectiveness and accessibility versus top-tier performance and intelligence. The presenter recommends Deepseek for users seeking a free and capable model, while suggesting Claude for those prioritizing the best possible performance and features, regardless of cost.