Use
COMPOSIOFOR YOU
Give Claude, Codex, or any harness secure access to 1500+ tools
CLI
One command to add tools and auth to your coding agent
Build
DEVELOPER PLATFORM
SDKs, tools, and auth for agents and backend apps
MCP GATEWAY
One managed MCP endpoint for all your tools and agents
ENTERPRISE
SSO, governance, and audit controls for deploying agents org-wide
Features
TRIGGERS
Start background agents from events across your connected apps
FIREWALLNEW
Blocks instructions hidden in tool results before your agent reads them
INSTANTNEW
Audio, video, search, browser and data tools, with no keys to manage
OPTIMIZATION
Lower cost per completed task with optimised tools and skills
FOR AGENTS
CLAUDE CODEX OPENCLAW CURSOR HERMES
BUILT FOR
OFFICE WORK SALES MARKETING PRODUCT & DESIGN CUSTOMER SUPPORT ENGINEERING HR & RECRUITING FINANCE & OPS E-COMMERCE CONTENT & MEDIA IT & SECURITY
ALL USE CASES
DOCS

Claude 3.7 Sonnet vs. Grok 3 vs. o3-mini-high

by ShrijalFeb 27, 202529 min read

ClaudeLLM

Just a week after Grok’s release, we now have the Claude 3.7 Sonnet, which certainly has eaten into Grok’s hype pie. Grok was definitely one of the best models for coding, and now, with the new Claude, the equations might change.
Anthropic has been clear about where to position its model: to become the best at coding, for the developers, by the developers, of the developers.

However, I have been using Grok 3 and o3-mini-high, which have performed well on most coding tasks. So, I wanted to see how the new Sonnet compares with Grok 3 and OpenAI’s o3-mini-high.

This is in no way a thorough benchmark; it’s mostly a vibe check, but it’s enough to gauge each model’s coding potential.
So, let’s dive in.

TL;DR

If you want to jump straight to the conclusion, when compared against these three models, Claude 3.7 Sonnet is the clear winner when writing code.

Grok 3 and o3-mini-high are somewhat similar, but if I had to compare them, Grok 3 would generate slightly better code than the o3-mini-high model.

Brief on Claude 3.7 Sonnet

This AI Model was released a few days ago and is already the talk of the tech community. I’m emphasizing tech because this model is widely regarded as the best AI model for code, at least for now.

Claude 3.7 Sonnet supports up to 128K output tokens (beta), which is over 15x longer than before. This is especially handy when generating longer and high-quality code.

It’s eating all the other AI models on the SWE Benchmark with an average accuracy of 62.3%. Its accuracy can even exceed 70%, the highest of any AI model. This represents a 13–20% accuracy gap compared to top OpenAI models, the previous Anthropic model Claude 3.5, and open-source models like DeepSeek R1.

Despite this power, Claude 3.7 has an 18% reduction in total costs compared to its earlier models. It maintains consistent token-based pricing at $3 per million input tokens and $15 per million output tokens.

Not only that, but Anthropic released Claude Code along with it, an agentic AI CLI that understands your codebase, helps fix issues, answers questions, and, with its Git integration, gives you an idea of your project history.

Check out this video to get a brief understanding of it:

Comparing Coding Abilities

I have high hopes for the Claude 3.7 Sonnet. Let’s see if we can spot any significant differences in code between Sonnet and the other two models.

Let’s start with something interesting:

1. Build simple Minecraft using Pygame

The task is simple: all three LLMs are asked to build a simple Minecraft game using Pygame.

Prompt: Build me a very simple Minecraft game using Pygame in Python.

Response from Claude 3.7 Sonnet

Here’s the code it generated:

import random
import sys

<p>import pygame</p>
<h1>Initialize Pygame</h1>
<p>pygame.init()</p>
<h1>Game settings</h1>
<p>SCREEN_WIDTH = 800
SCREEN_HEIGHT = 600
BLOCK_SIZE = 40
GRAVITY = 0.5
JUMP_STRENGTH = 10
PLAYER_SPEED = 5</p>
<h1>Colors</h1>
<p>SKY_COLOR = (135, 206, 235)
PLAYER_COLOR = (255, 0, 0)
GRASS_COLOR = (34, 139, 34)
DIRT_COLOR = (139, 69, 19)
STONE_COLOR = (105, 105, 105)
COAL_COLOR = (47, 47, 47)
INVENTORY_BG = (64, 64, 64, 128)</p>
<h1>Create screen</h1>
<p>screen = pygame.display.set_mode((SCREEN_WIDTH, SCREEN_HEIGHT))
pygame.display.set_caption("Simple Minecraft")
clock = pygame.time.Clock()</p>
<h1>Block types</h1>
<p>EMPTY = 0
GRASS = 1
DIRT = 2
STONE = 3
COAL = 4</p>
<h1>Block colors</h1>
<p>block_colors = {
    GRASS: GRASS_COLOR,
    DIRT: DIRT_COLOR,
    STONE: STONE_COLOR,
    COAL: COAL_COLOR,
}</p>
<h1>Define player</h1>
<p>class Player:
    def <strong>init</strong>(self, x, y):
        self.rect = pygame.Rect(x, y, BLOCK_SIZE - 10, int(BLOCK_SIZE * 1.5))
        self.velocity_y = 0
        self.on_ground = False
        self.selected_block = GRASS</p>
<pre><code>def move(self, dx, dy, blocks):
    # Move horizontally
    self.rect.x += dx
    self.check_collision(dx, 0, blocks)

    # Apply gravity
    self.velocity_y += GRAVITY
    self.rect.y += int(self.velocity_y)
    self.on_ground = False
    self.check_collision(0, self.velocity_y, blocks)

    # Keep player on screen
    if self.rect.left &lt; 0:
        self.rect.left = 0
    if self.rect.right &gt; SCREEN_WIDTH:
        self.rect.right = SCREEN_WIDTH

def check_collision(self, dx, dy, blocks):
    for y, row in enumerate(blocks):
        for x, block in enumerate(row):
            if block != EMPTY:
                block_rect = pygame.Rect(
                    x * BLOCK_SIZE, y * BLOCK_SIZE, BLOCK_SIZE, BLOCK_SIZE
                )
                if self.rect.colliderect(block_rect):
                    if dx &gt; 0:  # Moving right
                        self.rect.right = block_rect.left
                    if dx &lt; 0:  # Moving left
                        self.rect.left = block_rect.right
                    if dy &gt; 0:  # Moving down
                        self.rect.bottom = block_rect.top
                        self.on_ground = True
                        self.velocity_y = 0
                    if dy &lt; 0:  # Moving up
                        self.rect.top = block_rect.bottom
                        self.velocity_y = 0

def jump(self):
    if self.on_ground:
        self.velocity_y = -JUMP_STRENGTH

def draw(self, surface):
    pygame.draw.rect(surface, PLAYER_COLOR, self.rect)

Create world

def create_world(width, height): world = [[EMPTY for _ in range(width)] for _ in range(height)]

# Generate terrain
ground_height = height - 5

for x in range(width):
    height_variation = random.randint(-1, 1)
    ground_height += height_variation

    # Keep within bounds
    if ground_height &lt; height - 10:
        ground_height = height - 10
    if ground_height &gt; height - 3:
        ground_height = height - 3

    # Add grass at the top
    world[ground_height][x] = GRASS

    # Add dirt below grass
    for y in range(ground_height + 1, ground_height + 4):
        if y &lt; height:
            world[y][x] = DIRT

    # Add stone below dirt
    for y in range(ground_height + 4, height):
        if y &lt; height:
            if random.random() &lt; 0.1:  # 10% chance for coal
                world[y][x] = COAL
            else:
                world[y][x] = STONE

return world

Initialize game

world_width = SCREEN_WIDTH // BLOCK_SIZE world_height = SCREEN_HEIGHT // BLOCK_SIZE world = create_world(world_width, world_height) player = Player(SCREEN_WIDTH // 2, 0)

Game loop

running = True while running: for event in pygame.event.get(): if event.type == pygame.QUIT: running = False

    # Jump on space
    if event.type == pygame.KEYDOWN:
        if event.key == pygame.K_SPACE:
            player.jump()

        # Switch selected block
        if event.key == pygame.K_1:
            player.selected_block = GRASS
        elif event.key == pygame.K_2:
            player.selected_block = DIRT
        elif event.key == pygame.K_3:
            player.selected_block = STONE
        elif event.key == pygame.K_4:
            player.selected_block = COAL

    # Place or break blocks with mouse
    if event.type == pygame.MOUSEBUTTONDOWN:
        mouse_x, mouse_y = pygame.mouse.get_pos()
        block_x = mouse_x // BLOCK_SIZE
        block_y = mouse_y // BLOCK_SIZE

        # Make sure it's within world bounds
        if 0 &lt;= block_x &lt; world_width and 0 &lt;= block_y &lt; world_height:
            # Left click to place block
            if event.button == 1:
                # Don't place if player is there
                block_rect = pygame.Rect(
                    block_x * BLOCK_SIZE,
                    block_y * BLOCK_SIZE,
                    BLOCK_SIZE,
                    BLOCK_SIZE,
                )
                if not player.rect.colliderect(block_rect):
                    world[block_y][block_x] = player.selected_block

            # Right click to break block
            elif event.button == 3:
                world[block_y][block_x] = EMPTY

# Handle movement
keys = pygame.key.get_pressed()
dx = 0
if keys[pygame.K_a] or keys[pygame.K_LEFT]:
    dx = -PLAYER_SPEED
if keys[pygame.K_d] or keys[pygame.K_RIGHT]:
    dx = PLAYER_SPEED

player.move(dx, 0, world)

# Draw everything
screen.fill(SKY_COLOR)

# Draw blocks
for y, row in enumerate(world):
    for x, block in enumerate(row):
        if block != EMPTY:
            pygame.draw.rect(
                screen,
                block_colors[block],
                (x * BLOCK_SIZE, y * BLOCK_SIZE, BLOCK_SIZE, BLOCK_SIZE),
            )

# Draw player
player.draw(screen)

# Draw inventory
pygame.draw.rect(screen, INVENTORY_BG, (10, 10, 200, 50))
for i, block_type in enumerate([GRASS, DIRT, STONE, COAL]):
    block_rect = pygame.Rect(20 + i * 50, 15, 40, 40)
    pygame.draw.rect(screen, block_colors[block_type], block_rect)

    # Highlight selected block
    if block_type == player.selected_block:
        pygame.draw.rect(screen, (255, 255, 255), block_rect, 2)

    # Draw number
    font = pygame.font.SysFont(None, 24)
    text = font.render(str(i + 1), True, (255, 255, 255))
    screen.blit(text, (block_rect.x + 5, block_rect.y + 5))

# Update display
pygame.display.flip()
clock.tick(60)

pygame.quit() sys.exit()

Here’s the output of the program:

Response from Grok 3

Here’s the code it generated:

import pygame<p></p>
<h1>Initialize Pygame</h1>
<p>pygame.init()</p>
<h1>Screen settings</h1>
<p>WIDTH = 800
HEIGHT = 600
BLOCK_SIZE = 40
screen = pygame.display.set_mode((WIDTH, HEIGHT))
pygame.display.set_caption("Simple Minecraft 2D")</p>
<h1>Colors</h1>
<p>WHITE = (255, 255, 255)
BLACK = (0, 0, 0)
BROWN = (139, 69, 19)
GREEN = (0, 255, 0)
BLUE = (0, 0, 255)</p>
<h1>Player class</h1>
<p>class Player:
    def <strong>init</strong>(self, x, y):
        self.x = x
        self.y = y
        self.width = BLOCK_SIZE
        self.height = BLOCK_SIZE
        self.speed = 5</p>
<pre><code>def draw(self):
    pygame.draw.rect(screen, BLUE, (self.x, self.y, self.width, self.height))

Block class

class Block: def init(self, x, y, block_type): self.x = x self.y = y self.width = BLOCK_SIZE self.height = BLOCK_SIZE self.type = block_type

def draw(self):
    if self.type == "dirt":
        pygame.draw.rect(screen, BROWN, (self.x, self.y, self.width, self.height))
    elif self.type == "grass":
        pygame.draw.rect(screen, GREEN, (self.x, self.y, self.width, self.height))

Create game objects

player = Player(WIDTH // 2, HEIGHT // 2) blocks = []

Generate initial terrain

for x in range(0, WIDTH, BLOCK_SIZE): for y in range(HEIGHT // 2, HEIGHT, BLOCK_SIZE): if y == HEIGHT // 2: blocks.append(Block(x, y, "grass")) else: blocks.append(Block(x, y, "dirt"))

Game loop

clock = pygame.time.Clock() running = True inventory = []

while running: for event in pygame.event.get(): if event.type == pygame.QUIT: running = False elif event.type == pygame.MOUSEBUTTONDOWN: mouse_x, mouse_y = pygame.mouse.get_pos() # Left click to break block if event.button == 1: for block in blocks[:]: if ( block.x <= mouse_x < block.x + BLOCK_SIZE and block.y <= mouse_y < block.y + BLOCK_SIZE ): inventory.append(block.type) blocks.remove(block) break # Right click to place block elif event.button == 3 and inventory: grid_x = (mouse_x // BLOCK_SIZE) * BLOCK_SIZE grid_y = (mouse_y // BLOCK_SIZE) * BLOCK_SIZE # Check if space is empty occupied = False for block in blocks: if block.x == grid_x and block.y == grid_y: occupied = True break if not occupied and grid_y >= HEIGHT // 2: blocks.append(Block(grid_x, grid_y, inventory.pop()))

# Player movement
keys = pygame.key.get_pressed()
if keys[pygame.K_LEFT] and player.x &gt; 0:
    player.x -= player.speed
if keys[pygame.K_RIGHT] and player.x &lt; WIDTH - player.width:
    player.x += player.speed
if keys[pygame.K_UP] and player.y &gt; 0:
    player.y -= player.speed
if keys[pygame.K_DOWN] and player.y &lt; HEIGHT - player.height:
    player.y += player.speed

# Draw everything
screen.fill(WHITE)
for block in blocks:
    block.draw()
player.draw()

# Display inventory
font = pygame.font.Font(None, 36)
text = font.render(f"Inventory: {len(inventory)} blocks", True, BLACK)
screen.blit(text, (10, 10))

pygame.display.flip()
clock.tick(60)

pygame.quit()

Here’s the output of the program:

This program we got is pretty disappointing and definitely not what I expected. Nothing works as it should, except player movement. It looks way less like Minecraft and more like a snake game.

Response from OpenAI o3-mini-high

Here’s the code it generated:

import pygame<p></p>
<h1>Initialize Pygame</h1>
<p>pygame.init()</p>
<h1>Screen settings</h1>
<p>WIDTH = 800
HEIGHT = 600
BLOCK_SIZE = 40
screen = pygame.display.set_mode((WIDTH, HEIGHT))
pygame.display.set_caption("Simple Minecraft 2D")</p>
<h1>Colors</h1>
<p>WHITE = (255, 255, 255)
BLACK = (0, 0, 0)
BROWN = (139, 69, 19)
GREEN = (0, 255, 0)
BLUE = (0, 0, 255)</p>
<h1>Player class</h1>
<p>class Player:
    def <strong>init</strong>(self, x, y):
        self.x = x
        self.y = y
        self.width = BLOCK_SIZE
        self.height = BLOCK_SIZE
        self.speed = 5</p>
<pre><code>def draw(self):
    pygame.draw.rect(screen, BLUE, (self.x, self.y, self.width, self.height))

Block class

class Block: def init(self, x, y, block_type): self.x = x self.y = y self.width = BLOCK_SIZE self.height = BLOCK_SIZE self.type = block_type

def draw(self):
    if self.type == "dirt":
        pygame.draw.rect(screen, BROWN, (self.x, self.y, self.width, self.height))
    elif self.type == "grass":
        pygame.draw.rect(screen, GREEN, (self.x, self.y, self.width, self.height))

Create game objects

player = Player(WIDTH // 2, HEIGHT // 2) blocks = []

Generate initial terrain

for x in range(0, WIDTH, BLOCK_SIZE): for y in range(HEIGHT // 2, HEIGHT, BLOCK_SIZE): if y == HEIGHT // 2: blocks.append(Block(x, y, "grass")) else: blocks.append(Block(x, y, "dirt"))

Game loop

clock = pygame.time.Clock() running = True inventory = []

# Player movement
keys = pygame.key.get_pressed()
if keys[pygame.K_LEFT] and player.x &gt; 0:
    player.x -= player.speed
if keys[pygame.K_RIGHT] and player.x &lt; WIDTH - player.width:
    player.x += player.speed
if keys[pygame.K_UP] and player.y &gt; 0:
    player.y -= player.speed
if keys[pygame.K_DOWN] and player.y &lt; HEIGHT - player.height:
    player.y += player.speed

# Draw everything
screen.fill(WHITE)
for block in blocks:
    block.draw()
player.draw()

# Display inventory
font = pygame.font.Font(None, 36)
text = font.render(f"Inventory: {len(inventory)} blocks", True, BLACK)
screen.blit(text, (10, 10))

pygame.display.flip()
clock.tick(60)

pygame.quit()

Here’s the output of the program:

The output by the o3-mini-high model is pretty disappointing; we only got a blank screen with the added background colour.

Final Verdict: It’s pretty fair to say Claude 3.7 wins by a huge margin. Everything works just as expected. The overall game it built includes almost all the features I thought of.

2. Multiple Balls in a Spinning Hexagon

Let’s quickly test all these models with a standard question for judging different LLMs.

This is a modified version of the question where only one ball is spun up inside a spinning hexagon.

Prompt: Write me a Python script for 10 balls inside a fast-spinning hexagon.

Response from Claude 3.7 Sonnet

Here’s the code it generated:

import math
import random
import sys<p></p>
<p>import pygame</p>
<h1>Initialize pygame</h1>
<p>pygame.init()</p>
<h1>Constants</h1>
<p>WIDTH, HEIGHT = 800, 600
FPS = 60
WHITE = (255, 255, 255)
BLACK = (0, 0, 0)
GRAY = (100, 100, 100)
BALL_COLORS = [
    (255, 0, 0),  # Red
    (0, 255, 0),  # Green
    (0, 0, 255),  # Blue
    (255, 255, 0),  # Yellow
    (255, 0, 255),  # Magenta
    (0, 255, 255),  # Cyan
    (255, 165, 0),  # Orange
    (128, 0, 128),  # Purple
    (255, 192, 203),  # Pink
    (0, 128, 0),  # Dark Green
]</p>
<h1>Screen setup</h1>
<p>screen = pygame.display.set_mode((WIDTH, HEIGHT))
pygame.display.set_caption("Multiple Balls in a Fast-Spinning Hexagon")
clock = pygame.time.Clock()</p>
<h1>Hexagon properties</h1>
<p>hex_radius = 180
hex_center = (WIDTH // 2, HEIGHT // 2)
hex_rotation = 0
hex_rotation_speed = 2.0  # Increased speed (degrees per frame)</p>
<h1>Ball properties</h1>
<p>num_balls = 10
ball_radius = 15
balls = []</p>
<h1>Create balls with random positions and velocities</h1>
<p>for i in range(num_balls):
    # Random position inside the hexagon (with some margin from the edges)
    angle = random.uniform(0, 2 * math.pi)
    distance = random.uniform(0, hex_radius - ball_radius * 2)
    x = hex_center[0] + distance * math.cos(angle)
    y = hex_center[1] + distance * math.sin(angle)</p>
<pre><code># Random velocity
velocity_x = random.uniform(-3, 3)
velocity_y = random.uniform(-3, 3)

balls.append(
    {
        "pos": [x, y],
        "velocity": [velocity_x, velocity_y],
        "color": BALL_COLORS[i],
        "radius": ball_radius,
    }
)

Function to calculate hexagon vertices

def get_hexagon_vertices(center, radius, rotation): vertices = [] for i in range(6): angle_deg = rotation + 60 * i angle_rad = math.radians(angle_deg) x = center[0] + radius * math.cos(angle_rad) y = center[1] + radius * math.sin(angle_rad) vertices.append((x, y)) return vertices

Function to check collision with a line segment

def check_line_collision(ball_pos, ball_radius, line_start, line_end): x1, y1 = line_start x2, y2 = line_end x0, y0 = ball_pos

# Vector from line start to end
dx, dy = x2 - x1, y2 - y1

# Vector from line start to ball
dx0, dy0 = x0 - x1, y0 - y1

# Calculate projection length
t = max(0, min(1, (dx0 * dx + dy0 * dy) / (dx * dx + dy * dy)))

# Calculate closest point on the line
closest_x = x1 + t * dx
closest_y = y1 + t * dy

# Distance from ball to closest point
distance = math.sqrt((x0 - closest_x) ** 2 + (y0 - closest_y) ** 2)

# Return collision info
if distance &lt;= ball_radius:
    # Normal vector for reflection
    if distance &gt; 0:  # Avoid division by zero
        nx = (x0 - closest_x) / distance
        ny = (y0 - closest_y) / distance
    else:
        nx, ny = 0, -1  # Default normal if distance is zero

    return True, (nx, ny)
return False, None

Function to check collision between balls

def check_ball_collision(ball1, ball2): x1, y1 = ball1["pos"] x2, y2 = ball2["pos"]

# Distance between balls
distance = math.sqrt((x2 - x1) ** 2 + (y2 - y1) ** 2)

# If balls are colliding
if distance &lt;= (ball1["radius"] + ball2["radius"]):
    # Normal vector
    if distance &gt; 0:
        nx = (x2 - x1) / distance
        ny = (y2 - y1) / distance
    else:
        nx, ny = 0, 1  # Default if balls are at the same position

    return True, (nx, ny)
return False, None

Main game loop

gravity = 0.1 ball_speed_limit = 6 running = True

while running: for event in pygame.event.get(): if event.type == pygame.QUIT: running = False

# Update hexagon rotation
hex_rotation += hex_rotation_speed
if hex_rotation &gt;= 360:
    hex_rotation -= 360

# Get hexagon vertices
hex_vertices = get_hexagon_vertices(hex_center, hex_radius, hex_rotation)

# Update balls
for ball in balls:
    # Apply gravity
    ball["velocity"][1] += gravity

    # Update ball position
    ball["pos"][0] += ball["velocity"][0]
    ball["pos"][1] += ball["velocity"][1]

    # Check collision with hexagon sides
    for i in range(6):
        start = hex_vertices[i]
        end = hex_vertices[(i + 1) % 6]

        collision, normal = check_line_collision(
            ball["pos"], ball["radius"], start, end
        )
        if collision:
            # Reflect velocity across the normal
            nx, ny = normal
            dot_product = ball["velocity"][0] * nx + ball["velocity"][1] * ny
            ball["velocity"][0] -= 2 * dot_product * nx
            ball["velocity"][1] -= 2 * dot_product * ny

            # Apply some energy loss
            ball["velocity"][0] *= 0.9
            ball["velocity"][1] *= 0.9

            # Move ball away from the wall to avoid getting stuck
            ball["pos"][0] += nx * 2
            ball["pos"][1] += ny * 2

    # Limit ball speed
    speed = math.sqrt(ball["velocity"][0] ** 2 + ball["velocity"][1] ** 2)
    if speed &gt; ball_speed_limit:
        ball["velocity"][0] = (ball["velocity"][0] / speed) * ball_speed_limit
        ball["velocity"][1] = (ball["velocity"][1] / speed) * ball_speed_limit

# Check collisions between balls
for i in range(num_balls):
    for j in range(i + 1, num_balls):
        collision, normal = check_ball_collision(balls[i], balls[j])
        if collision:
            # Calculate relative velocity
            nx, ny = normal
            vx1, vy1 = balls[i]["velocity"]
            vx2, vy2 = balls[j]["velocity"]

            # Relative velocity along the normal
            vrn = (vx1 - vx2) * nx + (vy1 - vy2) * ny

            # If balls are moving away from each other, skip
            if vrn &lt; 0:
                continue

            # Mass (assume equal mass)
            m1 = 1.0
            m2 = 1.0

            # Elastic collision response
            impulse = 2.0 * vrn / (m1 + m2)

            # Update velocities
            balls[i]["velocity"][0] -= impulse * m2 * nx
            balls[i]["velocity"][1] -= impulse * m2 * ny
            balls[j]["velocity"][0] += impulse * m1 * nx
            balls[j]["velocity"][1] += impulse * m1 * ny

            # Move balls apart to avoid sticking
            overlap = (
                balls[i]["radius"]
                + balls[j]["radius"]
                - math.sqrt(
                    (balls[i]["pos"][0] - balls[j]["pos"][0]) ** 2
                    + (balls[i]["pos"][1] - balls[j]["pos"][1]) ** 2
                )
            )

            if overlap &gt; 0:
                # Move proportional to their velocities
                balls[i]["pos"][0] -= nx * overlap * 0.5
                balls[i]["pos"][1] -= ny * overlap * 0.5
                balls[j]["pos"][0] += nx * overlap * 0.5
                balls[j]["pos"][1] += ny * overlap * 0.5

# Draw everything
screen.fill(BLACK)

# Draw hexagon
pygame.draw.polygon(screen, GRAY, hex_vertices, 3)

# Draw balls
for ball in balls:
    pygame.draw.circle(
        screen,
        ball["color"],
        (int(ball["pos"][0]), int(ball["pos"][1])),
        ball["radius"],
    )

pygame.display.flip()
clock.tick(FPS)

pygame.quit() sys.exit()

The code has just a few minor issues, like how we are extracting values from the normal returned by the check_ball_collision method without checking if the value is None.

Other than that, everything looks quite good.

Here’s the output of the program:

import math
import random<p></p>
<p>import pygame</p>
<h1>Initialize Pygame</h1>
<p>pygame.init()</p>
<h1>Set up the display</h1>
<p>WIDTH = 800
HEIGHT = 600
screen = pygame.display.set_mode((WIDTH, HEIGHT))
pygame.display.set_caption("Balls in Spinning Hexagon")</p>
<h1>Colors</h1>
<p>WHITE = (255, 255, 255)
RED = (255, 0, 0)
BLUE = (0, 0, 255)</p>
<h1>Hexagon properties</h1>
<p>center_x = WIDTH // 2
center_y = HEIGHT // 2
hex_size = 200
hex_angle = 0
hex_speed = 0.05  # radians per frame</p>
<h1>Ball properties</h1>
<p>class Ball:
    def <strong>init</strong>(self):
        self.radius = 10
        self.x = center_x
        self.y = center_y
        self.dx = random.uniform(-3, 3)
        self.dy = random.uniform(-3, 3)
        self.color = (
            random.randint(0, 255),
            random.randint(0, 255),
            random.randint(0, 255),
        )</p>
<h1>Create balls</h1>
<p>balls = [Ball() for _ in range(10)]</p>
<h1>Get hexagon vertices</h1>
<p>def get_hexagon_vertices(x, y, size, angle):
    vertices = []
    for i in range(6):
        vertex_angle = angle + (math.pi / 3) * i
        vx = x + size * math.cos(vertex_angle)
        vy = y + size * math.sin(vertex_angle)
        vertices.append((vx, vy))
    return vertices</p>
<h1>Check collision with hexagon walls</h1>
<p>def check_collision(ball, vertices):
    for i in range(6):
        x1, y1 = vertices[i]
        x2, y2 = vertices[(i + 1) % 6]</p>
<pre><code>    # Line equation: ax + by + c = 0
    a = y2 - y1
    b = x1 - x2
    c = -(a * x1 + b * y1)

    # Distance from point to line
    distance = abs(a * ball.x + b * ball.y + c) / math.sqrt(a * a + b * b)

    if distance &lt;= ball.radius:
        # Normal vector
        nx = b
        ny = -a
        norm_length = math.sqrt(nx * nx + ny * ny)
        nx /= norm_length
        ny /= norm_length

        # Reflect velocity
        dot = ball.dx * nx + ball.dy * ny
        ball.dx = ball.dx - 2 * dot * nx
        ball.dy = ball.dy - 2 * dot * ny

        # Push ball away from wall
        penetration = ball.radius - distance
        ball.x += nx * penetration
        ball.y += ny * penetration

Main game loop

clock = pygame.time.Clock() running = True

while running: for event in pygame.event.get(): if event.type == pygame.QUIT: running = False

# Clear screen
screen.fill((0, 0, 0))

# Update hexagon rotation
hex_angle += hex_speed
vertices = get_hexagon_vertices(center_x, center_y, hex_size, hex_angle)

# Update balls
for ball in balls:
    # Move ball
    ball.x += ball.dx
    ball.y += ball.dy

    # Check collision with hexagon
    check_collision(ball, vertices)

    # Draw ball
    pygame.draw.circle(screen, ball.color, (int(ball.x), int(ball.y)), ball.radius)

# Draw hexagon
pygame.draw.polygon(screen, WHITE, vertices, 2)

# Update display
pygame.display.flip()
clock.tick(60)

pygame.quit()

There’s an issue: the balls are not supposed to be outside the hexagon.

When I asked it the standard question of one ball inside the spinning hexagon, it answered easily, but it couldn’t handle it when I tweaked it just a bit.

It is fair to say that this model could not handle this question well.

Response from OpenAI o3-mini-high

Here’s the generated code:

#!/usr/bin/env python3<p></p>
<p>import math
import random
import pygame</p>
<p>def get_hexagon_vertices(center, radius, angle_offset):
    """
    Returns a list of 6 pygame.Vector2 vertices for a regular hexagon.</p>
<pre><code>center      -- pygame.Vector2 at the center.
radius      -- Distance from center to a vertex.
angle_offset-- Overall rotation in radians.
"""
vertices = []
for i in range(6):
    # Each vertex is 60 degrees (pi/3 radians) apart.
    angle = math.radians(60 * i) + angle_offset
    vertex = center + pygame.Vector2(math.cos(angle) * radius,
                                     math.sin(angle) * radius)
    vertices.append(vertex)
return vertices

def point_in_poly(x, y, poly): """ Determines if point (x, y) is inside the polygon. poly is a list of (x, y) tuples. (Uses the ray-casting algorithm.) """ inside = False n = len(poly) p1x, p1y = poly[0] for i in range(n+1): p2x, p2y = poly[i % n] if (y > min(p1y, p2y)) and (y <= max(p1y, p2y)) and (x <= max(p1x, p2x)): if p1y != p2y: xinters = (y - p1y) * (p2x - p1x) / (p2y - p1y) + p1x else: xinters = p1x if p1x == p2x or x <= xinters: inside = not inside p1x, p1y = p2x, p2y return inside

def main(): pygame.init()

# Window size and title
width, height = 800, 600
screen = pygame.display.set_mode((width, height))
pygame.display.set_caption("10 Balls Inside a Fast Spinning Hexagon")
clock = pygame.time.Clock()

hex_center = pygame.Vector2(width/2, height/2)
hex_radius = 200           # Distance from center to vertex.
rotation_angle = 0         # Current rotation angle in radians.
rotation_speed = 5         # Radians per second (fast spinning).

ball_radius = 10
n_balls = 10
balls = []   # Each ball is a dictionary with keys: pos, vel, color.

# Use the hexagon's initial (unrotated) vertices for placing balls.
initial_vertices = get_hexagon_vertices(hex_center, hex_radius, rotation_angle)
# Convert vertices to a list of (x,y) tuples for point_in_poly testing.
initial_poly = [(v.x, v.y) for v in initial_vertices]

# Create n_balls with random positions INSIDE the hexagon and random velocities.
while len(balls) &lt; n_balls:
    test_x = random.uniform(hex_center.x - hex_radius, hex_center.x + hex_radius)
    test_y = random.uniform(hex_center.y - hex_radius, hex_center.y + hex_radius)
    # Only accept if inside the polygon:
    if point_in_poly(test_x, test_y, initial_poly):
        ball = {
            'pos': pygame.Vector2(test_x, test_y),
            # Random velocity (pixels per second). Adjust the numbers for speed.
            'vel': pygame.Vector2(random.uniform(-200, 200),
                                  random.uniform(-200, 200)),
            'color': (random.randint(50,255), random.randint(50,255), random.randint(50,255))
        }
        balls.append(ball)

running = True
while running:
    dt = clock.tick(60) / 1000.0   # Delta-time in seconds.

    for event in pygame.event.get():
        if event.type == pygame.QUIT:
            running = False
        # Optionally press ESC to quit.
        elif event.type == pygame.KEYDOWN and event.key == pygame.K_ESCAPE:
            running = False

    # Increase the rotation angle.
    rotation_angle += rotation_speed * dt
    # (Keep angle bounded to 2pi if you like; not strictly necessary.)
    rotation_angle %= (2 * math.pi)

    # Compute the current hexagon vertices.
    hex_vertices = get_hexagon_vertices(hex_center, hex_radius, rotation_angle)

    # For collision computations we need the “inward” normal for each edge.
    # For a given edge from A to B, the inward normal is chosen so that:
    #    n = N/|N|, where N = (-(B.y-A.y), (B.x-A.x)) or its negative,
    # and it must point towards the hexagon center.
    edge_normals = []
    n_edges = len(hex_vertices)
    for i in range(n_edges):
        A = hex_vertices[i]
        B = hex_vertices[(i+1) % n_edges]
        edge = B - A
        # Candidate normal (rotate 90 degrees)
        candidate = pygame.Vector2(-edge.y, edge.x)
        # Check – choose the normal that points toward the hexagon’s center.
        if candidate.dot(hex_center - A) &lt; 0:
            candidate = -candidate
        n = candidate.normalize()
        edge_normals.append((A, n))

    # Update each ball’s position and check for collision with each edge.
    for ball in balls:
        ball['pos'] += ball['vel'] * dt

        # Check collision with every edge of the hexagon.
        for A, n in edge_normals:
            # Compute how far the ball's center is along the inward normal from edge A.
            # (The signed distance.)
            d = (ball['pos'] - A).dot(n)
            if d &lt; ball_radius:
                # Only reflect if the ball is moving against the wall.
                if ball['vel'].dot(n) &lt; 0:
                    ball['vel'] = ball['vel'] - 2 * ball['vel'].dot(n) * n
                # Push the ball out of penetration.
                ball['pos'] += (ball_radius - d) * n

    screen.fill((0, 0, 0))  # clear with black.

    # Draw the hexagon (outline only).
    # Convert hexagon vertices from Vector2 to tuples.
    hex_points = [(int(v.x), int(v.y)) for v in hex_vertices]
    pygame.draw.polygon(screen, (255, 255, 255), hex_points, 3)

    # Draw balls.
    for ball in balls:
        pos = (int(ball['pos'].x), int(ball['pos'].y))
        pygame.draw.circle(screen, ball['color'], pos, ball_radius)

    pygame.display.flip()

pygame.quit()

if name == "main": main()

Final Verdict: On this question, I was pretty surprised that, except for the Grok 3 model, both Claude 3.7 and o3-mini-high got the answer correct. o3-mini-high seems to have outperformed Grok 3 here. However, Sonnet just blew them out of the water.

3. Build a real-time Browser-Based Markdown Editor

Considering how good Claude has been in both the tests of building a game and animation, and again with their solid claims on coding, let’s do a quick web app test on all three models.

Prompt: Build a tiny browser-based Markdown editor with syntax highlighting, export-to-PDF functionality, and a minimal UI using Tailwind in Next.js, with all the changes in a single file.

Response from Claude 3.7 Sonnet

"use client";<p></p>
<p>import { useState, useEffect } from "react";
import { marked } from "marked";
import hljs from "highlight.js";
import "highlight.js/styles/github.css";
import { jsPDF } from "jspdf";
import html2canvas from "html2canvas";</p>
<p>// Configure marked with highlight.js for syntax highlighting
marked.setOptions({
  highlight: function (code: string, lang: string) {
    if (lang && hljs.getLanguage(lang)) {
      return hljs.highlight(code, { language: lang }).value;
    }
    return hljs.highlightAuto(code).value;
  },
  breaks: true,
});</p>
<p>export default function MarkdownEditor() {
  const [markdown, setMarkdown] = useState<string>(
    '# Hello, Markdown!\\n\\nStart typing here...\\n\\n<code>javascript\\\\n// Code snippet example\\\\nfunction hello() {\\\\n  console.log("Hello, world!");\\\\n}\\\\n</code>',
  );
  const [html, setHtml] = useState<string>("");
  const [isDark, setIsDark] = useState<boolean>(false);</p>
<p>  // Convert markdown to HTML when input changes
  useEffect(() => {
    const convertMarkdown = async () => {
      const parsedHtml = await marked.parse(markdown);
      setHtml(parsedHtml);
    };</p>
<pre><code>convertMarkdown();

}, [markdown]);

// Toggle dark mode const toggleDarkMode = () => { setIsDark(!isDark); };

// Export to PDF const exportToPdf = async () => { const previewElement = document.getElementById("preview");

if (!previewElement) return;

// Set background to white for PDF export
const originalBackground = previewElement.style.background;
const originalColor = previewElement.style.color;
previewElement.style.background = "white";
previewElement.style.color = "black";
previewElement.style.padding = "20px";

try {
  const canvas = await html2canvas(previewElement, {
    scale: 2,
    useCORS: true,
    logging: false,
  });

  const imgData = canvas.toDataURL("image/png");
  const pdf = new jsPDF({
    orientation: "portrait",
    unit: "mm",
    format: "a4",
  });

  const pdfWidth = pdf.internal.pageSize.getWidth();
  const pdfHeight = pdf.internal.pageSize.getHeight();
  const imgWidth = canvas.width;
  const imgHeight = canvas.height;
  const ratio = Math.min(pdfWidth / imgWidth, pdfHeight / imgHeight);
  const imgX = (pdfWidth - imgWidth * ratio) / 2;
  const imgY = 30;

  pdf.addImage(
    imgData,
    "PNG",
    imgX,
    imgY,
    imgWidth * ratio,
    imgHeight * ratio,
  );
  pdf.save("markdown-document.pdf");
} finally {
  // Reset styling after export
  previewElement.style.background = originalBackground;
  previewElement.style.color = originalColor;
  previewElement.style.padding = "";
}

};

return ( <div className={min-h-screen ${isDark ? "bg-gray-900 text-white" : "bg-gray-100 text-gray-800"}} > {/* Header */} <header className={p-4 flex justify-between items-center ${isDark ? "bg-gray-800" : "bg-white border-b"}} > <h1 className="text-xl font-bold">Markdown Editor</h1> <div className="flex gap-2"> <button onClick={toggleDarkMode} className={px-3 py-1 rounded ${isDark ? "bg-gray-700 hover:bg-gray-600" : "bg-gray-200 hover:bg-gray-300"}} > {isDark ? "☀️ Light" : "🌙 Dark"} </button> <button onClick={exportToPdf} className="px-3 py-1 rounded bg-blue-500 text-white hover:bg-blue-600" > Export PDF </button> </div> </header>

  {/* Editor and Preview */}
  &lt;div className="flex flex-col md:flex-row h-[calc(100vh-64px)]"&gt;
    {/* Markdown Input */}
    &lt;div className="w-full md:w-1/2 h-1/2 md:h-full p-2"&gt;
      &lt;textarea
        className={`w-full h-full p-4 font-mono text-sm resize-none focus:outline-none rounded-md ${
          isDark
            ? "bg-gray-800 text-gray-100"
            : "bg-white border text-gray-800"
        }`}
        value={markdown}
        onChange={(e) =&gt; setMarkdown(e.target.value)}
        placeholder="Type markdown here..."
      /&gt;
    &lt;/div&gt;

    {/* Preview */}
    &lt;div className="w-full md:w-1/2 h-1/2 md:h-full p-2"&gt;
      &lt;div
        id="preview"
        className={`w-full h-full p-4 overflow-auto rounded-md prose ${
          isDark ? "bg-gray-800 prose-invert" : "bg-white border"
        } max-w-none`}
        dangerouslySetInnerHTML={{ __html: html }}
      /&gt;
    &lt;/div&gt;
  &lt;/div&gt;
&lt;/div&gt;

); }

The code is all good, except that code highlighting does not work correctly. This seems to be an issue because marked Recently introduced breaking changes to set up code highlighting.

The model may not be trained on the most recent data for the module.

Here’s the exported PDF:

Response from Grok 3

"use client";<p></p>
<p>import { useState } from "react";
import ReactMarkdown from "react-markdown";
import { Prism as SyntaxHighlighter } from "react-syntax-highlighter";
import { dracula } from "react-syntax-highlighter/dist/esm/styles/prism";
import { jsPDF } from "jspdf";</p>
<p>export default function MarkdownEditor() {
  const [content, setContent] = useState<string>(`# Welcome to Markdown Editor</p>
<p>Write some <strong>bold</strong> text or <em>italic</em> text here.</p>
<p>\\<code>\\\\</code>\\<code>javascript // Code example function hello() {   console.log("Hello, World!"); } \\\\</code>\\<code>\\\\</code>
`);</p>
<p>  const handleExportPDF = () => {
    const doc = new jsPDF();
    const pageWidth = doc.internal.pageSize.getWidth();
    const margin = 20;
    const maxWidth = pageWidth - 2 * margin;</p>
<pre><code>// Split content into lines and handle basic formatting
const lines = doc.splitTextToSize(content, maxWidth);
let y = margin;

lines.forEach((line: string) =&gt; {
  if (y &gt; 280) {
    doc.addPage();
    y = margin;
  }
  doc.text(line, margin, y);
  y += 7;
});

doc.save("markdown-export.pdf");

};

return ( <div className="min-h-screen bg-gray-100 flex flex-col"> <header className="bg-white shadow-sm p-4 flex justify-between items-center"> <h1 className="text-xl font-semibold text-gray-800">Markdown Editor</h1> <button onClick={handleExportPDF} className="px-4 py-2 bg-blue-600 text-white rounded hover:bg-blue-700 transition-colors" > Export to PDF </button> </header>

  &lt;main className="flex-1 flex p-6 gap-6 max-w-7xl mx-auto w-full"&gt;
    {/* Editor */}
    &lt;div className="flex-1"&gt;
      &lt;textarea
        value={content}
        onChange={(e) =&gt; setContent(e.target.value)}
        className="w-full h-[calc(100vh-120px)] p-4 bg-white rounded-lg shadow-sm border border-gray-200 focus:outline-none focus:ring-2 focus:ring-blue-500 font-mono text-sm resize-none"
        placeholder="Start writing markdown..."
      /&gt;
    &lt;/div&gt;

    {/* Preview */}
    &lt;div className="flex-1"&gt;
      &lt;div className="w-full h-[calc(100vh-120px)] p-4 bg-white rounded-lg shadow-sm border border-gray-200 overflow-auto"&gt;
        &lt;ReactMarkdown
          components={{
            code({ node, inline, className, children, ...props }) {
              const match = /language-(\\\\w+)/.exec(className || "");
              return !inline &amp;&amp; match ? (
                &lt;SyntaxHighlighter
                  style={dracula}
                  language={match[1]}
                  PreTag="div"
                  {...props}
                &gt;
                  {String(children).replace(/\\\\n$/, "")}
                &lt;/SyntaxHighlighter&gt;
              ) : (
                &lt;code className={className} {...props}&gt;
                  {children}
                &lt;/code&gt;
              );
            },
          }}
        &gt;
          {content}
        &lt;/ReactMarkdown&gt;
      &lt;/div&gt;
    &lt;/div&gt;
  &lt;/main&gt;
&lt;/div&gt;

); }

We have a couple of issues, the first one being that the headings don’t really work. There are also issues with the font contrast, and the exported PDF doesn’t render the Markdown correctly.

In the exported PDF, we get raw text instead of formatted Markdown, with no emoji support.

Response from OpenAI o3-mini-high

Here’s the generated code:

Verdict: Here, as well, Claude 3.7 is the clear winner compared to the other two models. Almost everything worked, but the other two models couldn’t get it right. There were some issues with text contrast and markdown rendering on the site and in the PDF.

4. Create a Code Diff Viewer

Let’s try a simple web application example to see if all of them get it right.

This is a pretty standard question and somewhat easy to implement. I have high hopes that all three of them should get it right (even o3-mini-high ). So far, I’m somewhat disappointed with it, though.

Prompt: Write a simple web application for a code diff viewer, a tool that takes two text inputs and highlights differences side by side.

Response from Claude 3.7 Sonnet

Response from Grok 3

Everything else seems to be working fine with this output as well, but in the diff, it is not taking the line’s indentation into account.

Response from OpenAI o3-mini-high

Here, we have an interesting result. It decided to use an external library diff and do highlighting per character and not per line.

Final Verdict: By functionality-wise, all three models got this problem correct (at least from testing at the surface level. There might be edge cases, though). I must say the overall code quality and output by o3-mini-high is comparatively better than both the Claude 3.7 Sonnet and Grok 3 model.

5. Manim code for Square to Pyramid Animation

Let’s end our test with a final Manim question. Most LLMs pretty much suck at writing Manim code, and it gets even harder when it’s a 3D scene.

Prompt: Create a Manim animation in Python where a 2D square smoothly lifts into 3D space to form a pyramid. Then, animate the pyramid shrinking back into a square.

Response from Claude 3.7 Sonnet

Response from Grok 3

Output

Response from OpenAI o3-mini-high

Here, this model really struggled with the 3D projection and failed to transform the square into a pyramid.

Final Verdict: The Claude 3.7 Sonnet and Grok 3 models got it completely correct, but in terms of animation, I prefer the output from the Grok 3 model. o3-mini-high failed completely and couldn’t even reach the solution.