Link the arXiv paper (2607.26504) in the model card

README.md

@@ -22,7 +22,8 @@ during generation** (a learned insertion head) and unmask tokens in any order,
 enabling genuinely any-order code generation.
 
 - Paper: *From Interface to Inference: Eliciting Any-Order Inference from Any-Order Models* —
-  S. Kim\*, J. Kim\*, T. Lee\*, Y. Chen\*, Y. Du, S. Kakade, S. Chen (preprint, 2026). <!-- FILL IN: paper URL -->
+  S. Kim\*, J. Kim\*, T. Lee\*, Y. Chen\*, Y. Du, S. Kakade, S. Chen.
+  [arXiv:2607.26504](https://arxiv.org/abs/2607.26504).
 - Code: [github.com/SeunggeunKimkr/genuine-any-order](https://github.com/SeunggeunKimkr/genuine-any-order), `FlexMDM/` subdirectory (training, data pipeline, inference, and evaluation).
 - Checkpoint: `global_step_49500` (inference weights; optimizer/RNG state stripped).