DeepSeek V3-0324: Generated 700 Lines Error-Free

DeepSeek V3 developed by the Chinese AI research lab DeepSeek under High-Flyer has been a standout in the AI landscape since its initial open-source release in December 2024. Known for its efficiency, performance, and accessibility, it continues to evolve rapidly. The latest update to DeepSeek V3, tagged “DeepSeek V3 0324” was rolled out on March 24, 2025, bringing subtle yet impactful refinements. Let’s look at these updates and try the new DeepSeek V3 model.

Minor Version Upgrade: DeepSeek V3 0324

• The upgrade enhances the user experience across DeepSeek’s official website, mobile app, and mini-program. This suggests a focus on streamlining interaction rather than altering core capabilities.
• The API interface and usage methods remain unchanged, ensuring continuity for developers. This means existing integrations (e.g., via model=’deepseek-chat’) don’t require adjustments.
• No major architectural changes were mentioned, indicating this is a refinement of the existing 671B-parameter Mixture-of-Experts (MoE) model, with 37B activated per token.
• Availability: The updated model is live on the official DeepSeek platforms (website, app, mini-program) and HuggingFace. The technical report and weights for “DeepSeek V3 0324” are accessible under the MIT license.

How is DeepSeek V3 0324 Performing?

DeepSeek V3 on Chatbot Arena leaderboard:

DeepSeek’s new V3 got a 55% score on a tough test (aider’s polyglot benchmark), which is much better than its previous version. Right now, it’s the second-best AI that doesn’t focus on deep thinking/reasoning, just behind Sonnet 3.7. It also performs close to more advanced thinking models like R1 and o3-mini.

A user on X tried the new DeepSeek V3 on his internal bench and it has a huge jump in all metrics on all tests. It is now the best non-reasoning model, dethroning Sonnet 3.5.

How to Access the Latest DeepSeek V3?

• Website: Test the updated V3 at deepseek.com for free.
• Mobile App: Available on iOS and Android, updated to reflect the March 24 release.
• API: Use model=’deepseek-chat’ at api-docs.deepseek.com. Pricing remains $0.14/million input tokens (promotional until February 8, 2025, though an extension hasn’t been ruled out).
• HuggingFace: Download the “DeepSeek V3 0324” weights and technical report from here.

Let’s Try the New DeepSeek V3 0324

I am going to use the updated DeepSeek model locally and via API.

Using DeepSeek-V3-0324 Locally with llm-mlx Plugin

Installation Steps

Here’s what you need to run it on your machine (assuming you’re using llm CLI + mlx backend):

!pip install llm
!llm install llm-mlx
!llm mlx download-model mlx-community/DeepSeek-V3-0324-4bit

This will:

1. Install the core llm CLI
2. Add the MLX backend plugin
3. Download the 4-bit quantized model (DeepSeek-V3-0324-4bit) — more memory-efficient

Run a Chat Prompt Locally

Example:

!llm chat -m mlx-community/DeepSeek-V3-0324-4bit 'Generate an SVG of a pelican riding a bicycle'

Output:

If the model runs successfully, it should respond with an SVG snippet of a pelican on a bike – goofy and glorious.

Using DeepSeek-V3-0324 via API

Install Required Package

!pip3 install openai

Yes, even though you’re using DeepSeek, you’re interfacing with it using OpenAI-compatible SDK syntax.

Python Script for API Interaction

Here’s a cleaned-up, annotated version of what’s happening in the script:

from openai import OpenAI
import time

# Timing setup
start_time = time.time()

# Initialize client with your DeepSeek API key and base URL
client = OpenAI(
    api_key="Your_api_key", 
    base_url="https://api.deepseek.com"  # This is important
)

# Send a streaming chat request
response = client.chat.completions.create(
    model="deepseek-chat",
    messages=[
        {"role": "system", "content": "You are a helpful assistant"},
        {"role": "user", "content": "How many r's are there in Strawberry"},
    ],
    stream=True
)

# Handle streamed response and collect metrics
prompt_tokens = 0
generated_tokens = 0
full_response = ""

for chunk in response:
    if hasattr(chunk, "usage") and hasattr(chunk.usage, "prompt_tokens"):
        prompt_tokens = chunk.usage.prompt_tokens

    if hasattr(chunk, "choices") and hasattr(chunk.choices[0], "delta") and hasattr(chunk.choices[0].delta, "content"):
        content = chunk.choices[0].delta.content
        if content:
            generated_tokens += 1
            full_response += content
            print(content, end="", flush=True)

# Performance tracking
end_time = time.time()
total_time = end_time - start_time

# Token/sec calculations
prompt_tps = prompt_tokens / total_time if prompt_tokens > 0 else 0
generation_tps = generated_tokens / total_time if generated_tokens > 0 else 0

# Output metrics
print("\n\n--- Performance Metrics ---")
print(f"Prompt: {prompt_tokens} tokens, {prompt_tps:.3f} tokens-per-sec")
print(f"Generation: {generated_tokens} tokens, {generation_tps:.3f} tokens-per-sec")
print(f"Total time: {total_time:.2f} seconds")
print(f"Full response length: {len(full_response)} characters")

Output

### Final Answer
After carefully examining each letter in "Strawberry," we find that the letter 'r' appears **3 times**.
**Answer:** There are **3 r's** in the word "Strawberry."
--- Performance Metrics ---
Prompt: 17 tokens, 0.709 tokens-per-sec
Generation: 576 tokens, 24.038 tokens-per-sec
Total time: 23.96 seconds
Full response length: 1923 characters

Find the full code and output here.

Building A Digital Marketing Website Using DeepSeek-V3-0324

Using DeepSeek-V3-0324, an advanced language model, to automatically generate a digital marketing landing page—modern, sleek, and small in scope—by using a prompt-based code generation approach.

!pip3 install openai

# Please install OpenAI SDK first: `pip3 install openai`

from openai import OpenAI
import time

# Record the start time
start_time = time.time()  # Add this line to initialize start_time

client = OpenAI(api_key="Your_API_KEY", base_url="https://api.deepseek.com")

response = client.chat.completions.create(
    model="deepseek-chat",
    messages=[
        {"role": "system", "content": "You are a Website Developer"},
        {"role": "user", "content": "Code a modern small digital marketing Landing page"},
    ],
    stream=True  # This line makes the response a stream of events
)

# Initialize variables to track tokens and content
prompt_tokens = 0
generated_tokens = 0
full_response = ""

# Process the stream
for chunk in response:
    # Track prompt tokens (usually only in first chunk)
    if hasattr(chunk, "usage") and hasattr(chunk.usage, "prompt_tokens"):
        prompt_tokens = chunk.usage.prompt_tokens

    # Track generated content
    if hasattr(chunk, "choices") and hasattr(chunk.choices[0], "delta") and hasattr(chunk.choices[0].delta, "content"):
        content = chunk.choices[0].delta.content
        if content:
            generated_tokens += 1
            full_response += content
            print(content, end="", flush=True)

# Calculate timing metrics
end_time = time.time()
total_time = end_time - start_time

# Calculate tokens per second
if prompt_tokens > 0:
    prompt_tps = prompt_tokens / total_time
else:
    prompt_tps = 0

if generated_tokens > 0:
    generation_tps = generated_tokens / total_time
else:
    generation_tps = 0

# Print metrics similar to the screenshot
print("\n\n--- Performance Metrics ---")
print(f"Prompt: {prompt_tokens} tokens, {prompt_tps:.3f} tokens-per-sec")
print(f"Generation: {generated_tokens} tokens, {generation_tps:.3f} tokens-per-sec")
print(f"Total time: {total_time:.2f} seconds")
print(f"Full response length: {len(full_response)} characters")

Output:

The page is for a digital marketing agency called “NexaGrowth” It uses a modern, clean design with a carefully chosen color palette The layout is responsive and uses contemporary web design techniques The navigation is fixed at the top of the page The hero section is designed to immediately capture attention with a large headline and call-to-action buttons.

You can view the website here.

Find the full code and output here.

Also Read:

Context from Older Updates (Post-December 2024 Baseline)

To clarify what’s new, here’s a quick recap of the V3 baseline before the March 24 update:

• Initial Release: DeepSeek V3 launched with 671B parameters, trained on 14.8T tokens for $5.5–$5.58M using 2.664M H800 GPU hours. It introduced Multi-Head Latent Attention (MLA), Multi-Token Prediction (MTP), and auxiliary-loss-free load balancing, achieving 60 tokens/second and outperforming Llama 3.1 405B.
• Post-Training: Reasoning capabilities from DeepSeek R1 were distilled into V3, enhancing its performance via Supervised Fine-Tuning (SFT) and Reinforcement Learning (RL), completed with just 0.124M additional GPU hours.
• The March update builds on this foundation, focusing on usability and targeted performance tweaks rather than a full overhaul.

Find all about DeepSeek V3 Frontier LLM, Trained on a $6M Budget

Conclusion

The DeepSeek V3 0324 update might seem small, but it brings big improvements. It’s faster now, handling tasks like math and coding quickly. It’s also very steady, giving good results every time, whether you’re coding or solving problems. Plus, it can write 700 lines of code without messing up, which is great for people who build things with code. It still uses the smart 671B-parameter setup and stays cheap to use. Try the new DeepSeek V3 0324 and tell me what you think in the comments!

Stay tuned to Analytics Vidhya Blog for more such content!

Hello, I am Nitika, a tech-savvy Content Creator and Marketer. Creativity and learning new things come naturally to me. I have expertise in creating result-driven content strategies. I am well versed in SEO Management, Keyword Operations, Web Content Writing, Communication, Content Strategy, Editing, and Writing.