elie @eliebakouch

If you’re a researcher working on RL, you should definitely try SmolLM3-3B and get another data point besides Qwen3-3B.

1) We didn’t have time to try RL during post training, so I think there’s still some room to build an even better version of smollm!

2) We released the