Welcome to Veyra AI

We build tiny English language models for fast local inference. We focus on compact, CPU-friendly models that are easy to run and fine-tune. Our work centres on general base models, function calling, Python-focused variants, distillation, RLVR, and benchmarks. Our vision is to have tiny models that can run on low-powered hardware while being able to reason and retrieve knowledge whether that be using tool calls or from memory, and it all be integrated into easy to use open source tools such as the Veyra CLI. Our models can be found on the Open SLM Leaderboard.

Models:

Note on Kairo models: These are experimental models used for architectural and dataset testing. They should not be treated as reliable general-purpose language models. None are currently available.

Benchmarks:

Note on benchmarks: Some of our benchmarks were created or expanded with AI assistance (e.g. generating candidate questions and answers).