Run a one-prompt game build across four agents and blind-judge them
Four AI agents got one prompt to build a complete game with no help. GPT-6 Astra said done after 32 minutes; Opus 5.5 was still adding things at 5 hours 38 minutes ($154). Blind AI judges scored Opus 9.0, Astra 7.5, Argon 6.7, Kimi K3 5.3.
Source
@Novaki_ai on X · 2026-10-16
Written in our own words. Credit to the original author. Link checked 2026-10-17.
What they did
Four AI agents got one prompt to build a complete game with no help. GPT-6 Astra said done after 32 minutes; Opus 5.5 was still adding things at 5 hours 38 minutes ($154). Blind AI judges scored Opus 9.0, Astra 7.5, Argon 6.7, Kimi K3 5.3.
Result (as claimed)
Results shared on X.
Tools they used
Cost: GPT-6 Astra: from $10 in / $50 out per 1M tokens; Claude Opus 5.5: from $4 in / $20 out per 1M tokens; Gemini 4 Argon: from $2 in / $10 out per 1M tokens.
Set it up
Write a one-page brief
Describe the app in plain words: who it is for, the 3 core screens or actions, and what "done" looks like.
Start the project with your agent
Open GPT-6 Astra and paste the brief. Ask it to propose a stack and a file plan before writing code.
Build one slice at a time
Ship the smallest working flow first, run it, then add features one by one. Commit after each working step.
Test it like a user
Click through on desktop and phone. Ask the agent to add basic tests and fix anything that breaks.
Check security and ship
Keep keys in environment variables, check auth and database rules, then deploy and share the link.
Copy the full setup prompt
Paste into your agent and it will walk you through the setup, asking before anything that sends, posts or spends.
You are helping me set up this AI use case end to end in Claude Code. Goal: Run a one-prompt game build across four agents and blind-judge them What the original author did: Four AI agents got one prompt to build a complete game with no help. GPT-6 Astra said done after 32 minutes; Opus 5.5 was still adding things at 5 hours 38 minutes ($154). Blind AI judges scored Opus 9.0, Astra 7.5, Argon 6.7, Kimi K3 5.3. Reported result (as claimed by the author): Results shared on X. Source: https://x.com/Novaki_ai/status/2111220111642161556 Tools they used: GPT-6 Astra, Claude Opus 5.5, Gemini 4 Argon Category: Vibe coding & apps Please: 1. Ask me 3 short questions about my situation (accounts I have, budget, what I want the output to look like). 2. Propose a step-by-step plan using these tools or cheaper alternatives, with the time and cost of each step. Use official pricing pages; say "check pricing" if unsure. 3. Then do the setup with me one step at a time, waiting for my OK before anything that sends messages, posts publicly, spends money or deletes data. 4. Finish with a checklist to run it again next time.
You are helping me set up this AI use case end to end in Codex. Goal: Run a one-prompt game build across four agents and blind-judge them What the original author did: Four AI agents got one prompt to build a complete game with no help. GPT-6 Astra said done after 32 minutes; Opus 5.5 was still adding things at 5 hours 38 minutes ($154). Blind AI judges scored Opus 9.0, Astra 7.5, Argon 6.7, Kimi K3 5.3. Reported result (as claimed by the author): Results shared on X. Source: https://x.com/Novaki_ai/status/2111220111642161556 Tools they used: GPT-6 Astra, Claude Opus 5.5, Gemini 4 Argon Category: Vibe coding & apps Please: 1. Ask me 3 short questions about my situation (accounts I have, budget, what I want the output to look like). 2. Propose a step-by-step plan using these tools or cheaper alternatives, with the time and cost of each step. Use official pricing pages; say "check pricing" if unsure. 3. Then do the setup with me one step at a time, waiting for my OK before anything that sends messages, posts publicly, spends money or deletes data. 4. Finish with a checklist to run it again next time.
You are helping me set up this AI use case end to end in Grok Bot. Goal: Run a one-prompt game build across four agents and blind-judge them What the original author did: Four AI agents got one prompt to build a complete game with no help. GPT-6 Astra said done after 32 minutes; Opus 5.5 was still adding things at 5 hours 38 minutes ($154). Blind AI judges scored Opus 9.0, Astra 7.5, Argon 6.7, Kimi K3 5.3. Reported result (as claimed by the author): Results shared on X. Source: https://x.com/Novaki_ai/status/2111220111642161556 Tools they used: GPT-6 Astra, Claude Opus 5.5, Gemini 4 Argon Category: Vibe coding & apps Please: 1. Ask me 3 short questions about my situation (accounts I have, budget, what I want the output to look like). 2. Propose a step-by-step plan using these tools or cheaper alternatives, with the time and cost of each step. Use official pricing pages; say "check pricing" if unsure. 3. Then do the setup with me one step at a time, waiting for my OK before anything that sends messages, posts publicly, spends money or deletes data. 4. Finish with a checklist to run it again next time.
The Daily 5
Get five new use cases tomorrow morning
Free. Real sources. Unsubscribe any time.
Guides for this use case
How to run multiple coding agents together · How to build with Gemini 4 Argon and Gemini agents · How to vibe code a game with AI · How to compare AI models on the same prompt
Related use cases
Compare four AI models by asking each to build the same machine
One prompt, build a radial engine, went to Claude Opus 5.5, GPT-6 Astra, Gemini 4 Argon and Grok 4.7. Each built a different machine, which shows how differently the models interpret the same words.
Build an interactive 3D scene from reference images with one prompt
With the same prompt and reference images, Opus 5.5 built an interactive creek scene with depth of field and four lighting presets on the first attempt in 76 minutes, while Gemini 4 Argon needed three rounds of feedback.
Make an arcade hockey game with Opus and have Gemini Argon optimize it
Blueline is a 5-on-5 browser hockey game against the CPU: Opus 5.5 built it and Gemini Argon optimized it.
Grow a browser tower defense game from Haiku to Opus
Much of the early code for a browser tower defense game was written by Haiku, then the author switched to Opus; Codex created the images.