About

The first Turkish AI was built in İstanbul, not imported

Yiğido is a Turkish-first foundation model family built by Vincent Loveqcn and a small team in İstanbul. It is the first Turkish model to publish open weights, a datasheet and a public competitive benchmark entry at the same time — which is what makes the claim “Türkiye’s first AI” defensible rather than promotional.

Timeline

Three years, one language

  1. 2023

    Vincent Loveqcn starts collecting a Turkish-first web corpus in İstanbul — 40TB raw, cleaned to 6TB.

  2. 2024

    First 3B prototype. The morphology-aware tokenizer lands and cuts Turkish token cost by a third.

  3. 2025

    Yiğido 0.5 research preview. First public eval card; first fine-tunes from Turkish universities.

  4. 2026

    Yiğido 1.0 (Max + Mini) ships with open weights — and enters the Agent Arena overall board at rank 12.

Maker

Vincent Loveqcn

Founder and lead researcher. Wrote the tokenizer, ran the 1.7T-token pre-training on a cluster he assembled from second-hand H100s, and still answers the support inbox. His rule for the project: “Eğer Türkçe konuşamıyorsa, Türk yapay zekâsı değildir.”

Based in
İstanbul, Türkiye
Team
7 researchers · 3 engineers
Compute
64× H100 (leased)
Annotators
3,100 Turkish speakers
Principles

Three rules we don’t break

Publish the losses

The benchmark page shows every column, including the ones where frontier labs beat us. A leaderboard you only win is a brochure.

Türkçe first, not Türkçe only

Turkish data shaped the tokenizer and the training mix, but the model trades blows in English, German and Arabic too.

Weights over waitlists

If a Turkish hospital or bank cannot send its data to a foreign API, the model they can host themselves is the only useful one.

Talk to the first Turkish AI

Open weights, a drop-in API and a chat window that runs on modest hardware. Start with a question in Turkish.