grok3_Benchmark.R

# AI Model Benchmark Scores Across Domains
# https://x.com/xai/status/1891699715298730482
library(tidyverse)

color_palette <- c(
  "Grok-3" = "#1b9e77",
  "Grok-3 mini" = "#66a61e",
  "Gemini-2 Pro" = "#e7298a",
  "DeepSeek-V3" = "#7570b3",
  "Claude 3.5 Sonnet" = "#d95f02",
  "GPT-4o" = "#e6ab02"
)

data <- tibble(
  Benchmark = rep(
    c("Math (AIME'24)", "Science (GPQA)", "Coding (LCB Oct-Feb)"),
    each = 6
  ),
  Model = rep(c("Grok-3", "Grok-3 mini", "Gemini-2 Pro", 
                "DeepSeek-V3", "Claude 3.5 Sonnet", "GPT-4o"), 3),
  Score = c(52, 40, 36, 40, 16, 9, # Math
            75, 65, 65, 59, 65, 50, # Science
            57, 41, 36, 40, 36, 34)   # Coding
)

data <- data %>%
  group_by(Benchmark) %>%
  mutate(Benchmark_Mean = mean(Score)) %>%
  ungroup() %>%
  arrange(desc(Benchmark_Mean), Benchmark, desc(Score)) %>%
  group_by(Benchmark) %>%
  mutate(Order = row_number()) %>%
  ungroup() %>%
  mutate(Benchmark = factor(Benchmark, 
                            levels = unique(Benchmark[order(-Benchmark_Mean)]))) %>%
  mutate(Model = factor(
    Model,
    levels = c(
      "Grok-3",
      "Grok-3 mini",
      "Gemini-2 Pro",
      "DeepSeek-V3",
      "Claude 3.5 Sonnet",
      "GPT-4o"
    )
  ))

ggplot(data, aes(x = Benchmark, y = Score, fill = Model, group = Order)) +
  geom_col(position = position_dodge(width = 0.9), width = 0.7) +
  geom_hline(yintercept = c(20, 40, 60), color = "black", linewidth = 0.5) +
  scale_fill_manual(values = color_palette, name = "AI Model") +
  labs(
    title = "AI Model Benchmark Scores Across Domains",
    subtitle = "Performance on Math, Science, and Coding Tasks, Feb 2025",
    x = "Benchmark Category",
    y = "Benchmark Score",
    caption = "Note:\nAIME'24 = American Invitational Mathematics Examination 2024\nGPQA = General-Purpose Question Answering\nLCB Oct-Feb = LeetCode Benchmark (Oct 2024 - Feb 2025)\n
Source: xAI Grok3 Launch Presentation, Feb 18, 2025"
  ) +
  theme(
    plot.background = element_rect(fill = "black", color = NA),
    panel.background = element_rect(fill = "black", color = NA),
    plot.margin = margin(20, 15, 20, 15),
    panel.grid.major = element_blank(),
    panel.grid.minor = element_blank(),
    axis.line.x = element_line(color = "gray30", linewidth = 0.5),
    axis.line.y = element_blank(),
    axis.text = element_text(color = "white", size = 12, family = "sans"),
    axis.title = element_text(color = "white", size = 14, family = "sans"),
    plot.title = element_text(hjust = 0.5, size = 16, 
                              face = "bold", color = "white", family = "sans"),
    plot.title.position = "plot",
    plot.caption.position = "plot",
    axis.ticks.length.y = unit(.2, "cm"),
    axis.ticks.y = element_line(colour = "gray30"),
    axis.ticks.length.x = unit(0, "cm"),
    plot.subtitle = element_text(hjust = 0.5, size = 12, 
                                 color = "white", family = "sans"),
    plot.caption = element_text(hjust = 0, size = 10, 
                                color = "gray90", family = "sans"),
    legend.position = "top",
    legend.background = element_blank(),
    legend.box.background = element_blank(),
    legend.text = element_text(color = "white", 
                               size = 12, family = "sans"),
    legend.title = element_text(color = "white", 
                                size = 14, face = "bold", family = "sans")
  ) +
  scale_y_continuous(limits = c(0, 80), 
                     breaks = c(0, 20, 40, 60, 80), expand = c(0, 0)) +
  geom_text(aes(label = round(Score, 1)), position = position_dodge(width = 0.9), 
  vjust = -0.3, size = 4, color = "white", family = "sans")

ggsave(
  "grok-3_Benchmark2025.png",
  dpi = 300,
  width = 7,
  height = 7
)