vectrayx-base

VectraYX-Base 260M

VectraYX-Base is a 260M-parameter Spanish cybersecurity language model trained from scratch using the same three-phase curriculum and replay-buffer recipe as presented in the paper VectraYX-Nano: A 42M-Parameter Spanish Cybersecurity Language Model with Curriculum Learning and Native Tool Use.

This model is scaled to a mid-tier architecture (d_model=1024, n_layers=16).

• Paper: VectraYX-Nano: A 42M-Parameter Spanish Cybersecurity Language Model
• Code: vectrayx/vectrayx-nano-paper
• Nano model: jsantillana/vectrayx-nano

---

Results (VectraYX-Bench, single seed)

B4=0.000 on mixed SFT is a corpus-density artifact — at ratio 1:21 (LoRA mini), Base reaches B4=0.445±0.201.

---

Architecture

The architecture follows the configuration defined in configs/base.json in the official repository.

---

Files

Training ran on AWS SageMaker ml.g5.xlarge (NVIDIA A10G 24GB), ~11 wall-clock hours, ~$11 USD.

---

Citation

@misc{santillana2026vectrayx,
  title     = {VectraYX-Nano: A 42M-Parameter Spanish Cybersecurity Language Model
               with Curriculum Learning and Native Tool Use},
  author    = {Santillana, Juan S.},
  year      = {2026},
  eprint    = {2605.13989},
  archivePrefix = {arXiv},
  primaryClass  = {cs.CL},
  url       = {https://arxiv.org/abs/2605.13989}
}