The Turkish AI & LLM Landscape in 2026: Who Is Building What

If you run an organisation in Türkiye and you are weighing an AI project, one question quietly shapes every other decision: whose model do you build on? For most of the last few years, the honest answer was "an American one, in English, adapted for Turkish as best we can." That is no longer the only answer. Türkiye now has a genuine ecosystem of home-grown large language models, telecom and retail giants training their own, and a national strategy pouring money into sovereign AI. This article maps who is building what, and—more usefully—what that means when you have a real project to ship.

We have kept this grounded. Where a fact is public and verifiable, we state it. Where the market is moving fast or a claim is fuzzy, we say so. Model versions and parameter counts change monthly; treat specific numbers as a snapshot of mid-2026, not gospel.

Why a "Turkish LLM" matters at all

A language model trained overwhelmingly on English text can speak Turkish, but it does so with a foreign accent you can feel in production. Turkish is agglutinative—meaning is stacked onto word roots through long chains of suffixes—so a single dictionary word can carry what English needs a whole clause to express. Models that never saw enough Turkish tokenise it inefficiently, misread morphology, and stumble on the cultural and legal context that business answers depend on.

For a consumer chatbot, that shows up as slightly-off phrasing. For an enterprise workflow—an assistant answering questions about a KVKK policy, a bank summarising a Turkish contract, a support agent handling Istanbul-dialect slang—it shows up as wrong answers delivered confidently. That gap is exactly what the Turkish-model ecosystem exists to close, and why it is worth understanding before you commit a budget.

The national and foundation models

The most striking development is that Turkish LLMs are now being built as public infrastructure, not just commercial products.

T3 AI (T3AI / "T3 AI'LE") is the flagship. Developed by the Turkish Technology Team (T3) Foundation together with defence firm Baykar, it launched a beta on the TEKNOFEST social platform and is open-source, positioned explicitly as an "ethical" national model. Its origin story is unusual: contributions from nearly 1,800 volunteers across 67 provinces, with partners including the Ministry of National Education, SETA, state broadcaster TRT, the Turkish Academy of Sciences, Anadolu Agency, and Microsoft. Whatever you make of the politics, T3 AI signals that Türkiye intends to own a foundational model rather than rent one.

TÜBİTAK's Bilge is the research-council entry—a domestically developed model from Türkiye's national science and technology body, part of a deliberate push toward sovereign cognitive infrastructure.

HAVELSAN's MAIN platform takes a different tack. Reported to run a roughly nine-billion-parameter model, MAIN is a defence-sector enterprise AI system designed to operate in secure, closed environments—the kind of air-gapped deployment that matters when data cannot leave the building. It is a preview of where regulated Turkish enterprises are heading: capable models that never touch a public cloud.

Sitting above all of this is government strategy. Türkiye has announced a multi-billion-dollar national AI push aimed at making the country a serious digital power by 2030, with instruments like AI vouchers for SMEs, dedicated AI growth zones, and validation labs. For a business, the practical takeaway is that building on Turkish AI is increasingly a supported direction, not a lonely bet.

Related: Self-Hosted LLMs for Enterprises: When Local Models Beat the Frontier

The commercial and enterprise models

Alongside the national effort, Türkiye's largest technology companies have trained models grounded in their own data and needs.

Turkcell built Turkcell-LLM-7b-v1, a Turkish-optimised model based on Mistral 7B and fine-tuned on roughly five billion Turkish tokens using efficient adaptation methods (LoRA/DoRA). As a telecom operator with vast Turkish-language customer interaction, Turkcell has both the data and the motive to make Turkish work properly—and it has published its work rather than hiding it.

Trendyol, the country's e-commerce heavyweight, has released a family of bilingual (Turkish/English) LLMs ranging from around 7B to 70B parameters, including specialised variants for domains such as cybersecurity. For a retail platform, an in-house model tuned on product, review, and support language is a direct competitive lever, not a science project.

VNGRS produced Kumru, one of the more talked-about efforts of the past year. Kumru is a Turkish LLM built largely from scratch—the widely discussed Kumru-2B launched open-source in October 2025 alongside the ~500 GB Turkish web corpus used to train it, with a larger model reportedly planned. "From scratch" matters: rather than bolting Turkish onto an English base, VNGRS trained on Turkish first, which tends to produce cleaner tokenisation and more natural output.

Earlier academic work like the TURNA model laid groundwork here, and the open-source Turkish NLP community—datasets, tokenisers, evaluation sets—remains the connective tissue that makes all of this possible.

A snapshot comparison

Model / project	Who	Character	Best mental model for
T3 AI	T3 Foundation + Baykar	Open-source, national, "ethical" framing	A sovereign default, community-backed
Bilge	TÜBİTAK	Research-council, sovereign infrastructure	State and public-sector alignment
MAIN	HAVELSAN	~9B, closed/secure environments	Air-gapped, regulated deployments
Turkcell-LLM-7b	Turkcell	Mistral-based, Turkish fine-tune, open	Telecom-grade Turkish, practical size
Trendyol LLMs	Trendyol	Bilingual 7B–70B, domain variants	Commercial, e-commerce-tuned scale
Kumru	VNGRS	Trained from scratch on Turkish	Cleanest native Turkish handling

Read the last column, not the parameter counts. The right model is the one whose character matches your constraint—regulation, language quality, deployment environment, or budget—far more than the one with the biggest number.

What about the global chatbots?

None of this means the global players have left the room. In practice most Turkish enterprises today still run production workloads on the frontier international models, because raw capability, tooling, and reliability remain ahead at the top end. That is a legitimate choice—especially for internal, lower-risk automation where the data is not sensitive.

The shift is that it is no longer the only choice, and increasingly not the automatic one. Three forces are pushing Turkish organisations toward local or hybrid stacks: data residency and KVKK (sending personal data to a foreign API is now a board-level question, not an engineering detail), cost at scale (once a workflow runs millions of calls, a right-sized local model can be dramatically cheaper), and language quality on genuinely Turkish tasks. The mature answer for most is not "local or global" but a deliberate mix—frontier models where capability wins, Turkish or self-hosted models where sovereignty, cost, or nuance win.

Related: GEO for Turkish Companies: How to Get Cited by ChatGPT and AI Search

What this means for your project

If you are choosing a foundation in 2026, a few practical principles hold regardless of which name is ascendant this quarter:

• Start from the constraint, not the model. Regulated data pushes you toward closed or self-hosted deployment (the MAIN pattern). High-volume automation pushes you toward cost-efficient local models. Nuanced Turkish content pushes you toward natively-trained models like Kumru or Turkcell's. Let the hardest constraint choose.
• Design for model portability. The ecosystem is moving too fast to marry one model for good. Build your workflows—retrieval, tools, guardrails, evaluation—so the underlying model is a swappable component, not a foundation poured in concrete.
• Evaluate on your own Turkish tasks. Public benchmarks rarely reflect your documents, your customers, or your dialect. A small, honest test set of real questions with known-good answers will tell you more than any leaderboard.
• Assume hybrid. The most robust architecture routes each task to the model that fits it—and keeps sensitive data on infrastructure you control while still reaching for frontier capability where it is safe to.
• Watch the incentives. Turkey's national push means Turkish models will keep improving and staying open. Building with the grain of that momentum is a reasonable long-term bet.

The takeaway

Türkiye in 2026 is no longer a market that simply consumes AI built elsewhere. It has national foundation models, telecom and retail giants training their own, a from-scratch challenger in Kumru, defence-grade closed-environment platforms, and state money behind all of it. For an organisation with a real project, that is genuinely good news: more options, more leverage, and a credible path to keeping Turkish data and Turkish nuance where they belong.

The catch is that more options make the design harder, not easier. Choosing well now means understanding your constraints before your model.

At Orbitra, we help Turkish organisations design AI automation and agentic workflows that stay portable across this shifting model landscape—matching frontier, local, and self-hosted models to the right task, and keeping sensitive data on infrastructure you control. If you are mapping where to build, we can help you turn this landscape into a concrete architecture and a first pilot.