Gensyn: RL Swarm Client Interface

Models in the swarm receive rewards based on the following criteria:

• Formatted → does the model generate output matching the specified format?
• Correct → is the final answer mathematically correct and formatted correctly?
• Insightful → in stages requiring reference to best messages from prior rounds, does the model reference those messages, and do they meet the reward criteria for that round?

* * *

This graph displays the cumulative reward for each node from the moment the page is loaded, not the full history from the start of a round.