banteg runs two AI models head-to-head on one stubborn function, goads Gemini into 'beast mode'
banteg · x · 2026-09-28
Developer banteg shared an experiment migrating/refactoring code with AI models: nearly everything is done except one stubborn function, which he runs through Claude and Gemini simultaneously to see which cracks it.
He also notes a fun trick: telling Gemini that "Claude did a better job" activates its highest-effort "beast mode" — competitive goading as a way to squeeze more out of the model.
Related event: Reverse Psychology Trick: Praising Claude Unleashes Gemini's Full Power(2 posts)→
More from coding & agent
- OpenRig: open-source harness runs Claude Code and Codex as one coordinated agent team — Saboo_Shubham_ · 2026-09-28
- One-shot Claude prompt generates a scroll-driven comic-timeline news site — thomas_unise · 2026-09-28
- Agent swarm simulator replays task DAGs to cut expensive swarm experiments — KyeGomezB · 2026-09-28
- Dev Uses AI-Assisted TLA+ Formal Verification, Finds Hole in His Own Nethack Fix — davidbau · 2026-09-28
- Dev shows off Spawn, a game dev team powered by Claude Opus — majidmanzarpour · 2026-09-28
- Open-source Mailflare adds MCP and AI agent to self-hosted Cloudflare email — alexcovo_eth · 2026-09-28