About Vaelico
The best open storytelling models exist. We serve them.
Vaelico builds and hosts frontier open models for interactive fiction and long-form narrative — the models writers actually want, served on the newest silicon.
The Gap
The strongest open models for storytelling — GLM-Steam, Iceblink — already exist. But the large inference providers focus on enterprise and coding workloads; creative use cases simply aren't their priority.
So the best open narrative models sit unserved on HuggingFace, while writers settle for proprietary APIs that refuse half their prompts. We close that gap: we take the best open storytelling models and serve them as a first-class, OpenAI-compatible API — at a price the hardware makes possible.
How We Build
Built on Blackwell.
Running natively in NVFP4 on NVIDIA Blackwell lets us serve 100B-parameter models at a per-token price that undercuts the proprietary labs — without the corporate guardrails.
- ▸ > Inference
- NVFP4 on NVIDIA Blackwell (B200)
- ▸ > Quality
- ~97% of full precision
- ▸ > Scale
- 100B+ MoE models, served efficiently
- ▸ > Serving
- vLLM / SGLang · OpenAI-compatible
- ▸ > Distribution
- Hosted API
Independent. Open. Building.
VAELICO STUDIO, S.L. is an independent company based in Spain, founded by Roger Salvador — one engineer, a training pipeline, and a roadmap. Our first models go live as our infrastructure comes online.