diff --git a/src/lighteval/models/base_model.py b/src/lighteval/models/base_model.py
index f1ba6151..8e69cdbb 100644
--- a/src/lighteval/models/base_model.py
+++ b/src/lighteval/models/base_model.py
@@ -554,7 +554,10 @@ def greedy_until(
                     # There will be truncation of at least one sample, maximum generation size will be one
                     max_new_tokens = 1
                 else:  # We can't allow generation of more than max_length
-                    max_new_tokens = min(self.max_length - context_size, max_new_tokens)
+                    if max_new_tokens is None:  # If generation size is not set, we go all the way
+                        max_new_tokens = self.max_length - context_size
+                    else:
+                        max_new_tokens = min(self.max_length - context_size, max_new_tokens)
 
                 prepared_batch = Batch(
                     input_ids=tokenized["input_ids"],