Just some months in the past, DeepSeek shook the AI world with its V3, R1, and Janus fashions. And just lately, Elon Musk’s Grok 3 has joined the social gathering! With the declare of being the neatest AI on Earth, expectations from Grok 3 are fairly excessive. As per the official benchmarks shared by the xAI staff on the launch occasion, it appears to be a monster mannequin, outperforming all its rivals on a number of benchmarks. However is it really ok for normal duties you and I do in our each day lives? Is it value spending $40 on an X Premium+ account to entry the mannequin? You’ll undoubtedly know the solutions by the tip of this text. Let’s start our competitors, testing Grok 3 vs DeepSeek R1 on 5 numerous prompts!
Activity 1: Python Code Era
Immediate: “Write a Python perform to generate a random maze utilizing depth-first search and visualize it utilizing Matplotlib”
Grok 3
import random
import numpy as np
import matplotlib.pyplot as plt
def generate_maze(width, peak):
"""
Generate a random maze utilizing Depth-First Search and visualize it with Matplotlib.
Parameters:
width (int): Width of the maze (variety of columns)
peak (int): Peak of the maze (variety of rows)
"""
# Initialize the maze grid with partitions (1) and paths (0)
maze = np.ones((peak * 2 + 1, width * 2 + 1), dtype=int)
maze[1::2, 1::2] = 0 # Set paths (odd rows/columns) to 0
def carve_path(x, y):
"""
Recursively carve paths within the maze utilizing DFS.
"""
instructions = [(0, 2), (2, 0), (-2, 0), (0, -2)] # Down, Proper, Up, Left
random.shuffle(instructions)
for dx, dy in instructions:
new_x, new_y = x + dx, y + dy
if (0 <= new_x < peak * 2 + 1 and
0 <= new_y < width * 2 + 1 and
maze[new_x, new_y] == 1):
maze[new_x - dx // 2, new_y - dy // 2] = 0 # Take away wall
maze[new_x, new_y] = 0 # Carve path
carve_path(new_x, new_y)
# Begin carving from the top-left path (1,1)
carve_path(1, 1)
# Guarantee entrance and exit are open (top-left and bottom-right paths)
maze[0, 1] = 0 # Entrance
maze[-1, -2] = 0 # Exit
# Visualize the maze
plt.determine(figsize=(width // 2, peak // 2))
plt.imshow(maze, cmap='binary', interpolation='nearest')
plt.title("Random Maze Generated with DFS")
plt.axis('off')
plt.present()
return maze
# Instance utilization
if __name__ == "__main__":
# Generate a 10x10 maze (precise grid shall be 21x21 because of partitions)
maze = generate_maze(10, 10)#import csvDeepSeek R1:
import random
import matplotlib.pyplot as plt
def generate_maze(width, peak):
# Initialize the grid with partitions (1)
grid = [[1 for _ in range(width)] for _ in vary(peak)]
# Start line
start_x, start_y = 1, 1
grid[start_y][start_x] = 0
# Stack for DFS
stack = [(start_x, start_y)]
# Instructions: up, proper, down, left
instructions = [(0, -2), (2, 0), (0, 2), (-2, 0)]
whereas stack:
x, y = stack[-1]
random.shuffle(instructions)
for dx, dy in instructions:
nx, ny = x + dx, y + dy
if 0 < nx < width - 1 and 0 < ny < peak - 1 and grid[ny][nx] == 1:
grid[ny][nx] = 0
grid[y + dy // 2][x + dx // 2] = 0
stack.append((nx, ny))
break
else:
stack.pop()
return grid
def plot_maze(grid):
peak = len(grid)
width = len(grid[0])
fig, ax = plt.subplots(figsize=(width, peak))
ax.set_xticks([])
ax.set_yticks([])
for y in vary(peak):
for x in vary(width):
if grid[y][x] == 1:
ax.add_patch(plt.Rectangle((x, peak - y - 1), 1, 1, shade="black"))
ax.set_xlim(0, width)
ax.set_ylim(0, peak)
ax.set_aspect('equal')
plt.gca().invert_yaxis()
plt.present()
# Parameters
maze_width = 21 # Ought to be odd
maze_height = 21 # Ought to be odd
# Generate and plot the maze
maze = generate_maze(maze_width, maze_height)
plot_maze(maze)#import csvOutput:
Commentary:
The DeepSeek R1 output presents a transparent, well-structured maze that seems professionally generated. The sharp and constant partitions create a correctly enclosed labyrinth, guaranteeing a difficult but visually clear pathfinding expertise. It adheres to good maze-generation rules, making it each practical and aesthetically pleasing.
In distinction, the Grok 3 output seems extremely pixelated, with much less distinguishable paths as a result of thick black-and-white grid construction. The general design feels extra chaotic, with pathways which might be much less outlined in comparison with DeepSeek R1. Whereas it does make the most of Depth-First Search (DFS) era, the visible readability and value are considerably inferior, making it a much less efficient maze illustration.
Verdict:
Grok 3 ❌ | DeepSeek R1 ✅
Activity 5: Internet Search
As DeepSeek R1 doesn’t enable internet search, I’ve used DeepSeek search choice for this activity!
Immediate: “What are the most recent developments in nuclear fusion know-how as of 2025? Fromat of ourput – Record of developments +supply”
Commentary:
Between DeepSeek R1 and Grok 3, DeepSeek R1 is the higher mannequin for answering research-heavy queries like developments in nuclear fusion know-how. The most important benefit of DeepSeek R1 is that it supplies direct, clickable supply hyperlinks for each declare, guaranteeing transparency and credibility. It additionally covers a wider vary of developments, together with fusion gasoline know-how, private-sector investments, regulatory modifications, and authorities initiatives, making its response extra complete. Moreover, its sources come from authoritative platforms corresponding to ITER, UKAEA, and the U.S. DOE, which considerably improves trustworthiness.
However, whereas Grok 3 delivers higher explanatory depth, it falls quick in credibility as a result of it doesn’t embody direct supply hyperlinks. As an alternative, it vaguely references information shops like BBC Information or MIT Information with out correct citations. A serious weak point is its reliance on posts from X (Twitter), which frequently include unverified claims. Whereas Grok 3 does acknowledge these limitations, the dearth of concrete sources makes it much less dependable for factual analysis.
Verdict:
Grok 3 ❌ | DeepSeek R1 ✅
Activity 3: Fundamental Animation Utilizing HTML
Immediate: “Create HTML + CSS code for a pink ball rotating inside a sq. constantly.”
Grok 3:
<!DOCTYPE html>
<html lang="en">
<head>
<meta charset="UTF-8">
<meta title="viewport" content material="width=device-width, initial-scale=1.0">
<title>Rotating Crimson Ball in Sq.</title>
<model>
physique {
margin: 0;
padding: 0;
peak: 100vh;
show: flex;
justify-content: heart;
align-items: heart;
background-color: #f0f0f0; /* Mild grey background for distinction */
overflow: hidden;
}
.container {
place: relative;
width: 300px;
peak: 300px;
background-color: #000; /* Black sq. */
border: 2px strong #333; /* Darkish border for visibility */
}
.ball {
place: absolute;
width: 30px;
peak: 30px;
background-color: #ff0000; /* Crimson ball */
border-radius: 50%; /* Makes it round */
animation: rotateBall 3s linear infinite; /* Steady rotation */
transform-origin: 150px 150px; /* Middle of the sq. (150px from left/high) */
}
@keyframes rotateBall {
0% {
remodel: rotate(0deg) translate(120px) rotate(0deg);
}
100% {
remodel: rotate(360deg) translate(120px) rotate(-360deg);
}
}
</model>
</head>
<physique>
<div class="container">
<div class="ball"></div>
</div>
</physique>
</html>DeepSeek R1:
<!DOCTYPE html>
<html>
<head>
<model>
.container {
width: 200px;
peak: 200px;
border: 2px strong #000;
margin: 50px auto;
place: relative;
overflow: hidden;
}
.observe {
width: 100%;
peak: 100%;
animation: rotate 4s linear infinite;
transform-origin: heart;
}
.ball {
width: 20px;
peak: 20px;
background-color: pink;
border-radius: 50%;
place: absolute;
high: 0;
left: 50%;
remodel: translateX(-50%);
}
@keyframes rotate {
from {
remodel: rotate(0deg);
}
to {
remodel: rotate(360deg);
}
}
</model>
</head>
<physique>
<div class="container">
<div class="observe">
<div class="ball"></div>
</div>
</div>
</physique>
</html>Output:
Commentary:
Clearly, Grok 3 is struggling to interpret the immediate and generate the right response. DeepSeek R1 took about 8-10 minutes to answer, however its output is extra exact and correct.
Verdict:
Grok 3 ❌ | DeepSeek R1 ✅
Activity 4: Picture Evaluation
Immediate: “Analyze this chessboard place. Recommend one of the best transfer for the present participant (white) to checkmate black and clarify the reasoning“
Grok 3:
DeepSeek R1:
Commentary:
Each Grok 3 and DeepSeek R1 acquired the reply fallacious. Grok 3 urged e4-e5, which doesn’t ship checkmate or threaten the king. DeepSeek R1 urged Qe1#, however this transfer is unattainable within the given place, exhibiting it misinterpreted the board. The right transfer was Qf7#, the place the queen delivers checkmate by trapping the Black king. Grok 3 failed to acknowledge a direct checkmate, whereas DeepSeek R1 assumed the fallacious board setup as an alternative of analyzing the precise place.
Verdict:
Grok 3 ❌ | DeepSeek R1 ❌
Activity 5: Logical Reasoning
Immediate: “Clear up this zebra puzzle. Give me output in a desk”
Grok 3:
Placing the generated response within the puzzle:
DeepSeek R1:
Placing the generated response within the puzzle:
Commentary:
DeepSeek R1 once more took an extended to reply however gave the right reply. Grok 3 failed to know the picture and gave incorrect output.
Verdict:
Grok 3 ❌ | DeepSeek R1 ✅
Grok 3 vs DeepSeek R1: End result
| Python Code Era | ✅ DeepSeek R1 |
| Internet Search | ✅ DeepSeek R1 |
| Fundamental Animation (HTML + CSS) | ✅ DeepSeek R1 |
| Picture Evaluation (Chessboard Checkmate) | ❌ Each Failed |
| Logical Reasoning (Zebra Puzzle) | ✅ DeepSeek R1 |
Additionally Learn:
Finish Observe
Elon Musk’s Grok 3 was hyped as a game-changer in AI, claiming to be the neatest mannequin on Earth. Nevertheless, in real-world testing, it didn’t dwell as much as the expectations. Throughout a number of duties, Grok 3 struggled with accuracy, logical reasoning, and complicated problem-solving, usually producing incorrect or poorly structured responses. In the meantime, DeepSeek R1 constantly outperformed it, delivering extra correct, structured, and verifiable solutions in key areas like code era, internet search, and logical reasoning.
Regardless of the daring advertising and marketing claims, Grok 3 nonetheless has an extended option to go earlier than it could possibly compete with high AI fashions. The truth that it failed fundamental reasoning duties means that xAI wants main enhancements in its coaching method. Nevertheless, given Musk’s observe document of speedy iteration and enhancements, will probably be attention-grabbing to see if future updates can bridge this hole. Will Grok 3 evolve into the AI powerhouse it claims to be, or will it stay an overhyped experiment? Time will inform.
Keep tuned to Analytics Vidhya Weblog to observe Grok 3 updates usually!
Hiya, I’m Nitika, a tech-savvy Content material Creator and Marketer. Creativity and studying new issues come naturally to me. I’ve experience in creating result-driven content material methods. I’m properly versed in website positioning Administration, Key phrase Operations, Internet Content material Writing, Communication, Content material Technique, Modifying, and Writing.
