How the models are related

Weights carried forwardSteering reused

RL means reinforcement learning. ❄ marks the base kept fixed during adapter training.

Faynt, Reyn and Reyn 2v2 with harness Reyn 2v2 with harness faynt → reyn4: Faynt weights reused. reyn4 → reyn26: New router; gameplay weights reused faynt → 2v2: Team rewards + policy fine-tuning reyn4 → 2v2: Steering adapters, refitted for 2v2. faynt → wombo: Combo-adapter RL + scripted-teacher guidance Steering, refitted Faynt (1v1): Human-replay imitation.; RL for fighting. Policy trained THE FIGHTING POLICY Faynt 1v1 Human-replay imitation. RL for fighting. Policy trained Reyn 4C (1v1): Labelled imitation.; Specialists → distillation.; Paraphrase router. Backbone frozen in RL DEVELOPMENT BASELINE Reyn 4C 1v1 Labelled imitation. Specialists → distillation. Paraphrase router. Backbone frozen in RL Reyn 26C (1v1): New text router.; Same gameplay weights. Backbone frozen CURRENT RELEASE Reyn 26C 1v1 New text router. Same gameplay weights. Backbone frozen Wombo Combo (2v1): Combo adapter + RL.; Teacher-guided imitation. Faynt stays frozen COMBO LAB · BONUS VIDEO Wombo Combo 2v1 Combo adapter + RL. Teacher-guided imitation. Faynt stays frozen Reyn Pair (2v2): Train a Fox pair.; Team-reward RL.; Refit steering adapters. Frozen for steering STEERABLE FOX PAIR Reyn Pair 2v2 Train a Fox pair. Team-reward RL. Refit steering adapters. Frozen for steering
Faynt, Reyn and Reyn 2v2 with harness Reyn 2v2 with harness faynt → reyn4: Faynt weights reused. reyn4 → reyn26: New router; gameplay weights reused faynt → 2v2: Team rewards + policy fine-tuning reyn4 → 2v2: Steering adapters, refitted for 2v2. faynt → wombo: Combo-adapter RL + scripted-teacher guidance Faynt (1v1): Human-replay imitation.; RL for fighting. Policy trained Fighting policy Faynt 1v1 Policy trained Reyn 4C (1v1): Labelled imitation.; Specialists → distillation.; Paraphrase router. Backbone frozen in RL 4C baseline Reyn 4C 1v1 Base frozen Reyn 26C (1v1): New text router.; Same gameplay weights. Backbone frozen Current release Reyn 26C 1v1 Base frozen Wombo Combo (2v1): Combo adapter + RL.; Teacher-guided imitation. Faynt stays frozen Bonus video · lab Wombo Combo 2v1 Faynt frozen Reyn Pair (2v2): Train a Fox pair.; Team-reward RL.; Refit steering adapters. Frozen for steering Steerable Fox pair Reyn Pair 2v2 Base frozen

Select a model for its training details.

C indicates character coverage. The final router update keeps the gameplay weights fixed. All parameters stay fixed during play.