We asked four AI coding agents to rebuild Minesweeper—the results were explosive \ stacker news ~AI

pull down to refresh

We asked four AI coding agents to rebuild Minesweeper—the results were explosive arstechnica.com/ai/2025/12/the-ars-technica-ai-coding-agent-test-minesweeper-edition/

266 sats \ 7 comments \ @jakoyoh629 19 Dec AI

Agent 1: Mistral Vibe

https://cdn.arstechnica.net/wp-content/uploads/2025/12/aimine1.png

Agent 2: OpenAI Codex

https://cdn.arstechnica.net/wp-content/uploads/2025/12/aimine2.png

Agent 3: Anthropic Claude Code

https://cdn.arstechnica.net/wp-content/uploads/2025/12/aimine3.png

Agent 4: Google Gemini CLI

https://cdn.arstechnica.net/wp-content/uploads/2025/12/aimine4.png

view all related items

157 sats \ 5 replies \ @k00b 19 Dec

Someone could start a pretty awesome website that only has agents perform practical tasks like this and compares them. It's the Techcrunch of tomorrow. Benchmarks leave me wanting.

146 sats \ 4 replies \ @optimism 19 Dec

https://lmarena.ai

RESULTSRESULTS

lambda-1201-2:

gemini-3-pro:

125 sats \ 1 reply \ @k00b 19 Dec

That's better than the anecdotes I imagined but less entertaining

0 sats \ 0 replies \ @optimism 19 Dec

Updated with results... both work. lol

100 sats \ 1 reply \ @jakoyoh629 OP 19 Dec

Source: https://github.com/nickarocho/minesweeper

Ahahah

20 sats \ 0 replies \ @optimism 19 Dec

Haha!

32 sats \ 0 replies \ @optimism 19 Dec

lol.