Announcement5 min read
Introducing Tomaris: The First Uzbek-Native AI
Today we're announcing Tomaris — a 27-billion-parameter language model aligned from the ground up for Uzbek, live at tomaris.ai.
July 1, 2026Read
Blog
Insights, research, and updates from the Tomaris team.
Research6 min read
Why We Built Our Own Uzbek Benchmark
You cannot improve what you cannot measure — and no public benchmark measures Uzbek. So we built a 300-prompt eval, including 105 Olympiad-grade problems.
June 18, 2026Read
Engineering8 min read
Building a 227-Million-Word Uzbek Corpus
How we collected, cleaned, and curated 227 million words of high-quality Uzbek text — and why the next 16,000 books matter even more.
June 4, 2026Read
Research7 min read
Why Low-Resource Languages Need Dedicated AI
Generic AI models fail at Uzbek in predictable, structural ways. Here's why fine-tuning a dedicated model beats waiting for the giants.
May 21, 2026Read
Product5 min read
Agents That Work in Uzbek
Task-specific agents — study help, legal drafting, business analysis — built on top of Tomaris, with reasoning you can inspect.
May 7, 2026Read
Policy6 min read
Data Sovereignty and AI in Central Asia
Every query sent to a foreign AI API is data leaving the country. Why data residency matters, and how Tomaris keeps intelligence local.
April 23, 2026Read