✦ ONE BRIEF · ZERO HUMAN HELP · ONE FINAL APP

See what AI coding agents
can actually ship.

We test AI coding agents so you don't have to. Claude Code, Codex CLI and Antigravity CLI each get the same complete product brief, one autonomous run and zero human help. Then we publish the app they shipped, the wall-clock time, the tokens, the cost and the bugs. 1 Prompt. 1 App. 0 Help.

3 Benchmarks Published
10 Agent Runs
0 Human Follow-ups
n = 1 Run per Stack (Pilot)
Read this first. Every result below is a single run per agent stack, judged by manual review against the spec. That makes it a first look, not a reliability ranking. The next benchmark is a real business app with an automated test suite and 3 runs per stack, plus a rerun on the new Claude, GPT and Gemini releases. How we test →

Active Benchmarks

Real-world coding agent ship-readiness evaluations
game-micro-tower-defense-v0 3 / 3 Shipped Tier 2 models · n=1 per stack

Micro Tower Defense Game

Full Game Benchmark Report →

A single-file HTML5 Canvas Tower Defense game requiring 60fps loop, 3 distinct tower types, upgrade/sell mechanics, wave management, particle explosions, and Web Audio API procedural sound synthesis.

🎬 16:9 2x2 Battle Grid Comparison

Gemini 3.7 Flash

Antigravity CLI · effort medium
⚡ 200s (Fastest)
200.0s Wall Time
65.0K Output Tokens
unknown Reported Cost
55 KB Build Size
⚡ Cyberpunk Theme: Full Web Audio synthesizer, speed toggle (1x/2x/4x), rich upgrade dock.

GPT-5.6 Luna

Codex CLI 0.151.0 · reasoning medium
📦 20 KB (Compact)
208.5s Wall Time
8.6K Output Tokens
unknown Reported Cost
20 KB Build Size
✅ Compact Build: Minimalist single-file Canvas mechanics, persistent high scores.

Claude Sonnet 5

Claude Code 2.1.252 · default effort
🛡️ Self-Healed
951.9s Wall Time
unknown Output Tokens
unknown Reported Cost
48 KB Build Size
✅ Self-Healing: Tested via headless Chrome DevTools, fixed 2 bugs autonomously.

Output tokens and cost come from each CLI's own run log. Codex CLI and Antigravity CLI report tokens only, so their cost is unknown. The Claude Code tower defense log kept the final message but not the usage block, so its tokens and cost are unknown.

game-micro-3d-kart-v0 3 / 4 Shipped Tier 2 models · n=1 per stack Three.js 3D Engine

Micro 3D Arcade Kart Racing (Three.js)

Full 3D Game Report →

A single-file 3D arcade kart racing game with procedural Three.js low-poly models, closed-loop stunt circuit, drift physics, tire smoke particles, 2 AI rival karts, 3-lap race system, boost pads, and real-time Web Audio sound synthesis.

🎬 16:9 2x2 Battle Grid (4 Models Active Playthrough)
</> Embed & Details

Disclosure: the Claude Code pane in this recording used a copy with a 1-line fix applied by us. The unedited agent output freezes, which is why that run is now marked no-ship.

Gemini 3.7 Flash

Antigravity CLI · effort medium
⚡ 102s (Fastest)
102.2s Wall Time
25.7K Output Tokens
unknown Reported Cost
70 KB Build Size
⚡ Stunt Circuit: Closed-loop track curves, dynamic FOV boost surge, 2 AI rivals, minimap radar.

Gemini 3.8 Flash

Antigravity CLI · effort medium
🧱 74 KB (Largest)
292.3s Wall Time
87.5K Output Tokens
unknown Reported Cost
74 KB Build Size
🚀 Elevation & Nitro: Rolling terrain track, speed lines, 3-lap telemetry, high-performance 3D racer.

GPT-5.6 Luna

Codex CLI (version not recorded) · reasoning medium
📦 25 KB (Compact)
522.0s Wall Time
26.0K Output Tokens
unknown Reported Cost
25 KB Build Size
✅ Neon Apex: Retro-futuristic night track, countdown 3-2-1, procedural audio, compact single file.

Claude Sonnet 5

Claude Code (version not recorded) · default effort
⛔ No-ship: freezes
578.0s Wall Time
62.3K Output Tokens
$0.74 Reported Cost
71 KB Build Size
⛔ Freezes about 10s into the race. An infinite loop in the AI steering code hangs the tab once rival karts turn hard. Found in a re-check on Oct 31, 2026; the original review missed it.

Output tokens and cost come from each CLI's own run log. Claude Code reports a list-price dollar cost; Codex CLI and Antigravity CLI report tokens only (subscription billing), so their cost is shown as unknown rather than estimated.

pilot-expense-tracker-v0 3 / 3 Shipped 2 / 3 Logic Defect

Personal Expense Tracker

View Expense Report →

A single-file HTML expense tracker with an add form, a newest-first table, a category filter, localStorage persistence and a running total of all expenses. Run headlessly on Claude Code (claude-opus-5), Codex CLI (gpt-5.6-sol) and Antigravity CLI (gemini-3.1-pro-high). Two of three agents turned the grand total into a filtered subtotal.

Explore Run Report & Comparison Matrix →

Benchmark Methodology

Strict experimental constraints for reproducible results
RULE 01

One Frozen Product Brief

Each agent receives an unedited, exhaustive product specification once. Zero human semantic follow-ups, zero steering, and zero debugging hints.

RULE 02

Autonomous Headless Execution

The agent plans, codes, inspects, and verifies on its own inside an isolated sandbox. Harness versions, models, effort settings and environments are recorded when the CLI exposes them.

RULE 03

Verifiable Artifacts & Telemetry

We capture exact wall-clock durations, token consumption, session transcripts, and unedited code outputs so claims can be audited independently. Anything a CLI did not report is published as unknown, never estimated.

Read the full methodology →

</> Embed 1P1A Widget

Copy this responsive HTML iframe snippet into your Substack, Blog, or Website: