# 1P1A > 1P1A (1 Prompt. 1 App. 0 Help.) is an independent benchmark of AI coding agents. Each agent stack gets one complete product brief, runs autonomously with zero human follow-up, and we publish the app it shipped plus wall-clock time, tokens, reported cost, defects and full stack metadata (model, settings, harness version, tool permissions, environment, task version, run number). Unknown metadata is published as unknown, never estimated. Current status (Oct 31, 2026): 3 benchmarks, 10 runs, one run per agent stack, verdicts by manual review. This is a first look, not a reliability ranking. 1P1A does not name a best AI coding agent yet. ## Results - [Claude Code vs Codex CLI vs Antigravity CLI](https://1p1a.com/compare/claude-code-vs-codex-vs-antigravity/): all 10 runs side by side. Ship rate: Codex CLI 3/3, Antigravity CLI 3/3, Claude Code 2/3. Antigravity CLI was fastest in all 3 tasks. Codex CLI was the only agent to implement the expense total exactly as specified. - [Tower defense game benchmark](https://1p1a.com/game-micro-tower-defense-v0/): single-file HTML5 canvas game. All 3 shipped. Wall time 200.0s (Antigravity CLI + gemini-3.7-flash-medium), 208.5s (Codex CLI 0.151.0 + gpt-5.6-luna), 951.9s (Claude Code 2.1.252 + claude-sonnet-5, which self-tested in headless Chrome and fixed 2 bugs). - [3D kart racing game benchmark](https://1p1a.com/game-micro-3d-kart-v0/): single-file Three.js game. 3 of 4 shipped. Claude Code + claude-sonnet-5 froze about 10 s into the race (infinite loop in AI steering); corrected to no-ship on Oct 31, 2026. - [Expense tracker app benchmark](https://1p1a.com/pilot-expense-tracker-v0/): 7-requirement spec. All 3 shipped. Claude Code (claude-opus-5) and Antigravity CLI (gemini-3.1-pro-high) turned the grand total into a filtered subtotal; Codex CLI (gpt-5.6-sol) kept it correct. ## Answers and data - [AI coding agents FAQ](https://1p1a.com/faq/): short sourced answers (best agent, fastest, Claude Code cost per app, spec accuracy, common bugs). - [runs.json](https://1p1a.com/data/runs.json): every run as structured data (stack, verdict, wall time, tokens, reported cost, notes). - [llms-full.txt](https://1p1a.com/llms-full.txt): all findings, every run and the full methodology in one plain-text file. ## Method - [Methodology](https://1p1a.com/methodology/): run rules, metrics, verdicts, cost policy, corrections, neutrality. - Cost: published only when the CLI reports it. Claude Code reports list-price cost; Codex CLI and Antigravity CLI report tokens only on subscription billing, so their cost is unknown. - Reliability claims require at least 3 runs per stack on the same frozen task. - Data license: CC BY 4.0, credit 1P1A with a link to https://1p1a.com/. ## Contact - [@tomcandev on X](https://x.com/tomcandev)