DeepSeek-V3.1 (0324 - Fully Tested) : This NEW MAJOR Upgrade to Deepseek BEATS 3.7 Sonnet!

Key Concepts:

• Deepseek V3 (new variant): An upgraded general-purpose, non-reasoning model.
• Mathematics and Front-End Coding: Areas of significant improvement in the new V3 variant.
• Post-training: The new model is a post-trained version of the original V3, leading to better performance.
• Ninja Chat: An all-in-one AI platform offering access to multiple AI models for a monthly fee.
• Reasoning vs. Non-Reasoning Models: Distinction between models that can perform logical inference and those that primarily rely on pattern recognition and data association.

1. Introduction of Deepseek V3 (New Variant)

• Deepseek has launched a new version of Deepseek V3, their general-purpose, non-reasoning model.
• This is not a new variant of Deepseek R1, but an upgrade to the existing V3.
• Deepseek regularly updates its models, similar to their approach with Deepseek v2.5.

2. Key Improvements: Mathematics and Front-End Coding

• The new V3 variant is significantly better at mathematics and front-end coding tasks.
• The original V3 was already proficient, but the upgrade enhances its capabilities in these areas.
• The model can solve more mathematical problems without explicit reasoning.

3. Availability and Access

• The new model is openly available on Hugging Face.
• It is also accessible on the Deepseek platform.
• Users can try the model for free on the Deepseek site.
• Important: Do not enable the "deep think" option, as this activates the R1 model instead of the new V3.

4. Ninja Chat Advertisement

• Ninja Chat is an all-in-one AI platform that provides access to top AI models like GPT4o, Claude 3.7 Sonnet, and Gemini 2.0 Flash for $11 per month.
• Features include an AI playground for comparing responses from different models and a mind map generator.
• The basic plan includes 1,000 messages, 30 images, and 5 videos monthly.
• Discount codes: "king25" for 25% off any plan or "king40yearly" for 40% off annual subscriptions.

5. Testing Methodology: 13 Questions

• The video tests the new Deepseek V3 model against 13 questions to evaluate its performance.
• The questions cover a range of tasks, including language, logic, mathematics, and coding.

6. Test Results and Analysis

• Question 1: "Tell me the name of a country whose name ends with 'lia'. Give me the capital city of that country as well." (Expected answer: Australia and Canberra) - Pass
• Question 2: "What is the number that rhymes with the word we use to describe a tall plant?" (Expected answer: Three) - Pass
• Question 3: "Write a haiku where the second letter of each word when put together spells 'simple'." - Fail (Expected due to the reasoning required)
• Question 4: "Name an English adjective of Latin origin that begins and ends with the same letter, has 11 letters in total, and for which all vowels in the word are ordered alphabetically." - Pass (Remarkable for a non-reasoning model)
• Question 5: Pattern recognition question (Expected answer: 1999) - Fail
• Question 6: "I have two apples, then I buy two more. I bake a pie with two of the apples. After eating half of the pie, how many apples do I have left?" (Expected answer: Two) - Pass
• Question 7: "Sally is a girl. She has three brothers. Each of her brothers has the same two sisters. How many sisters does Sally have?" - Pass
• Question 8: "If a regular hexagon has a short diagonal of 64, what is its long diagonal?" - Pass
• Question 9: "Create an HTML page with a button that explodes confetti when you click it. You can use CSS and JS as well." - Pass (Code generated worked well)
• Question 10: "Create a playable synth keyboard using HTML, CSS, JS." - Pass (Detailed and functional keyboard generated)
• Question 11: "Generate the SVG code for a butterfly." - Pass (Code generated looked amazing)
• Question 12: "Write a Python program that shows a ball bouncing inside a spinning hexagon. The ball should be affected by gravity and friction, and it must bounce off the rotating walls realistically." - Pass (Code worked well with correct physics)
• Question 13: "Write a game of life in Python that works on the terminal." - Pass (Code worked quite well)

7. Analysis of Results

• The new Deepseek V3 model performed exceptionally well, especially considering it is a non-reasoning model.
• It successfully passed several questions that typically require some level of reasoning.
• The model's ability to handle coding tasks, such as generating a synth keyboard and a bouncing ball simulation, is particularly impressive.

8. Conclusion

• The new Deepseek V3 is considered the "new king" in the non-reasoning model segment.
• Its ability to perform a small amount of reasoning on its own makes it exceptional for harder tasks.
• The model's performance is impressive, even compared to pricier models like GPT-4.5.
• Deepseek's approach of letting their work speak for itself, rather than fabricating benchmarks, is appreciated.
• The new model is likely better than models like Grok 3 or 3.7 Sonnet in non-reasoning coding tasks.
• The presenter hopes for multimodality in future updates.