Brief, mostly facts
LLM agents in 2v2 Pokemon Showdown: talking helps teammates win
A solo-author post reports a 64% win rate with free talk versus 42% without, against a max-damage bot. The code is on GitHub.
What the research found
Solo-author research post builds a 2v2 Pokemon Showdown multi-agent environment where Qwen3.8-27B teammates with free talk win 64% vs a max-damage bot, versus 42% with no talk; code is on GitHub.
Claims and how well the evidence supports them
- Teammates who could talk freely won 64% of battles against the max-damage bot 99%
- Teammates who could not talk won 42% 99%
- Evaluation uses Qwen3.8-27B in a 2v2 Pokemon Showdown format 98%
- Code is available on GitHub at OliverGrabner/multi-agent-poke-showdown 97%
Evidence
- olivergrabner.com https://olivergrabner.com/research/pokemon
- github.com https://github.com/OliverGrabner/multi-agent-poke-showdown
Actionability rated low.