Hands-On Review: Gemini 2.5 Pro (0605) — Coding, Logic & Creative Writing Benchmarked Against o3 and Claude 4

Google recently launched Gemini 2.5 Pro (version 0605), claiming dominance in reasoning, mathematics, science, and programming. To validate these assertions, I conducted rigorous testing, generating everything from polished data visualization reports to a fully interactive React virtual gallery—all powered by this model.

Here’s what you’ll discover in this review:
Coding Prowess: Real-world performance in front-end animation and complex app development.

Logic & Creativity: How Gemini 2.5 Pro stacks up against o3 in reasoning and narrative craftsmanship.

End-to-End Workflow: My iterative process for building a React app and data report.

Multi-Model Faceoff: Direct comparisons with o3 and Claude 4 across diverse tasks.




I. Code Capabilities: From Animations to Full Applications

Google touts Gemini 2.5 Pro’s top scores on WebDevArena, a blind-test platform where users anonymously vote on AI-generated front-end outputs. But benchmarks ≠ real-world utility. Here’s how it performed:
Front-End Animation: Where Art Meets Engineering

Dancing Terracotta Warriors:

While older models produced robotic “calisthenics,” Gemini 2.5 Pro generated fluid, breakdance-like sequences with dynamic floor moves.


Interactive 3D Menger Sponge (P5.js):

Created a responsive fractal explorer: mouse-click to refine detail, scroll to zoom, drag to rotate—all with smooth gradient rendering.
Infinite Stellar Voyage

o3 Output: Functional but lacked depth.

Gemini 2.5 Pro: Rendered a cinematic starfield with parallax layers and warp-speed effects—my personal favorite.



Claw Machine Simulator

Gemini’s version featured realistic physics, dynamic lighting, and intuitive joystick controls—outperforming rivals in interactivity polish.



II. Logic & Creativity: Clash of the AI Titans

Raw coding skill isn’t enough; elite models must master reasoning and storytelling.
Logic Test: “Who Won the Funding?”

Problem: Three teams (A/B/C) compete for grants under complex rules.

o3: Eliminated C logically but couldn’t resolve A vs. B—rigorous yet incomplete.

Gemini 2.5 Pro: Declared B the winner via meta-reasoning:

“Since logic puzzles must have unique solutions, and B is the only team without contradictions—B wins.”

Verdict: Clever but unsound. o3 demonstrated stricter deductive rigor.
Creative Writing: Lighthouse Keeper & Cat

Warm/Touching Style:

o3: Poetic and vivid (“amber-eyed cat,” “counting stars with seashell songs”).

Gemini: Coherent but flat—read like a children’s book draft.

Gothic/Horror Style:

o3: Chilling atmosphere (“Is the lighthouse a sanctuary or a tribunal?”).

Gemini: Forced eeriness (“I am both the eye’s warden and its captive”)—weaker emotional pull.



Conclusion: o3 dominates narrative depth and stylistic versatility.


III. Tabular Data → Visualization Report

Parsing complex tables remains a notorious LLM pain point.
Gemini 2.5 Pro: Instantly summarized multi-model benchmark tables with 98% accuracy—zero tooling required.

o3: Used Code Interpreter for Python processing but misinterpreted Gemini-2-Flash’s video capabilities after minutes of computation.



Edge: Gemini’s native multi-modal understanding enables swift, precise data digestion.


IV. Build Quest: React Virtual Gallery from Scratch

Goal: A React app where users:
Upload images → display in 3D gallery

Click artwork for details + AI voice narration

Development Saga:
Initial Hurdle: Gemini refused to use React Router v7, claiming it was “alpha”—required manual override.

Context Amnesia: Forgot core features (e.g., image uploads) mid-task.

Debugging Hell: Syntax errors, dependency clashes, and runtime crashes plagued every iteration.

Final Collapse: Suggested “reinstall Node.js” after failing to resolve state-management bugs.

Claude Sonnet 4 Rescue: Fixed all issues in 1/3 the time.

Reality Check: Gemini excels as a “snippet generator” but falters in multi-step project orchestration.


Conclusion: Strengths, Gaps & Verdict

✅ Strengths:
Elite Front-End Code: Stunning animations/UI components (e.g., 3D interactives).

Cost Efficiency: ~4.4x cheaper input tokens than o3; ideal for budget-sensitive workflows.

Multi-Modal Agility: Tables, docs → visualizations sans external tools.

⚠️ Weaknesses:
Logic Rigor: Falls short of o3’s structured reasoning.

Creative Nuance: Lacks o3’s lyrical depth and tonal control.

Project Scalability: Crashes under complex, context-heavy builds.

Final Thought: Gemini 2.5 Pro (0605) is a front-end virtuoso and data whisperer—but for mission-critical logic or end-to-end engineering, o3 remains the sherpa.


Tested on Google AI Studio (Free Tier) | All images generated/curated by Gemini 2.5 Pro