Announcements and updates
EVREN LLM inference service opens: 11 models behind one OpenAI-compatible API
Industry Agenda ·
EVREN LLM inference service opens: 11 models behind one OpenAI-compatible API
In an announcement on its LinkedIn page on 18 September 2026, the EVREN Platform opened the EVREN LLM inference service to all users. According to the announcement, a fleet of 11 open-weight models running on a domestic bare-metal NVIDIA H200 cluster is served behind a single OpenAI-compatible API. The news reached a wider audience on 23 September when the Savunma Kariyer (Defence Careers) page shared it.
The platform, introduced to academia through YÖK in June, had focused on computer vision until then; the announcement says this first phase is complete and that language models are the second step. This news piece summarises the scope of the service, how the infrastructure is organised and what to watch for when connecting it to enterprise applications. Aksiyon Soft has no partnership with SSB or EVREN; this is independent industry news.
In short
- What: the EVREN Platform opened its LLM inference service to all users on 18 September 2026.
- Fleet: 11 open-weight models in four categories, with a context window of up to 1 million tokens.
- Interface: one OpenAI-compatible API; existing apps change only the address and the key.
- Data: prompts, responses and usage logs are processed in Türkiye and not transferred abroad.
- Cost: no separate contract, application or invoice process; calls are not deducted from credits until 1 November 2026.
Which models are in the fleet?
The announcement groups the models into four categories. The table below lists categories and model names exactly as announced; the usage examples in the right-hand column are our commentary based on enterprise project experience.
| Category | Models | Example enterprise use |
|---|---|---|
| Reasoning · Code · Agents | glm-5.3, deepseek-v4-flash, qwen3.8-flash-next, gemma-4-31b | internal support assistant, code review help, multi-step workflow agents |
| Vision and Video | qwen3-vl-30b | interpreting field reports with photos, summarising visual content |
| Search · Embeddings · Safety | qwen3-embedding-8b, qwen3-reranker-8b, qwen3guard-4b | internal document search (RAG), result ranking, prompt and response safety filtering |
| Documents and Speech | dots-ocr, deepseek-ocr-2, qwen3-asr-1.7b | reading scanned documents and invoices, transcribing call recordings |
According to the announcement, the context window goes up to 1 million tokens, meaning a document hundreds of pages long can be processed in one go. The announcement does not break this limit down by model, so for long-document scenarios you should confirm the model choice against the information in the platform dashboard.
How is the infrastructure organised?
According to the announcement, the first phase, computer vision, is complete; model training continues on the same bare-metal cluster, while vision inference runs on a separate server. The LLM service uses a separate GPU pool carved out of the same cluster. In our reading, this separation is meant to keep vision workloads and language model workloads from directly eating into each other’s capacity.
The data commitment is clear: prompts, responses and usage logs are processed in Türkiye and not transferred abroad. For organisations that must assess cross-border transfers under KVKK, Türkiye’s data protection law, this directly shapes architecture decisions.
How does the OpenAI-compatible API connect to existing apps?
The most important sentence in the announcement for developers is that existing applications can connect by changing only the address and the key. In an app that uses OpenAI client libraries or frameworks supporting that contract, the base endpoint is replaced with the address shown in the platform dashboard and the key with one generated on the platform. One of the fleet’s model names is passed as the model.
Compatibility is at the contract level; each model may still behave differently with tool calls, structured output and streaming. Before switching, we recommend running a small comparison test with your own prompt set and reviewing the results side by side with your current provider’s output.
What does it mean for enterprise software teams?
The two issues that most often stall enterprise LLM projects are data residency and procurement. EVREN LLM addresses both directly: data is processed in Türkiye, there is no separate contract, application or invoice process, and credits earned on the platform also apply to LLMs. Calls not being deducted from credits until 1 November 2026 opens a low-risk window for a pilot.
In our assessment, “just change the address” is technically true but not enough for production. A control layer between the application and the service should mask personal data, enforce quotas per user and department, write every call to an audit log and be able to fail over to a backup model or provider when the service does not respond.
Pilot checklist
- Have you chosen a single use case and a measurable success criterion for the pilot?
- Are the address and key stored in configuration or a secrets vault rather than in application code?
- Is personal data in prompts masked before it is sent?
- Have you run a model comparison test with your own prompt set?
- Is there a budget and monitoring plan for credit consumption after 1 November?
- Is a backup model or provider defined for service outages?
EVREN series
This is the third instalment of our series following the EVREN platform. The series started with the release of EVREN SDK and continued in June when the platform opened to universities through YÖK. A few days later the national AI platform EVREN was officially launched for broad use. For the steps to connect the API to enterprise software, see our EVREN API integration guide.
How can Aksiyon Soft help?
Our Samsun-based team works remotely with organisations across Türkiye and makes planned on-site visits when needed. Through our API and integration service we build the control layer that manages LLM calls, masking and audit logging, and model routing and fallback rules. To connect several systems to the same AI services, our API and data integration platform solution provides a shared foundation.
The process starts with a discovery session, followed by a single-scenario MVP, two-week sprint demos and hypercare plus SLA-backed support after go-live. For architecture details, read our articles on the AI-ops layer and LLM router and taking AI agents to production.
Frequently asked questions
Who can use the EVREN LLM service?
According to the announcement, the service is open to all EVREN users. Sign-in is through e-Devlet, and there is no separate application process for the LLM service.
Is it paid?
Credits earned on the platform also apply to LLMs. According to the announcement, all calls made until 1 November 2026 are not deducted from credits, and there is no separate contract or invoice process.
Where do I get the API address?
The endpoint address and key are in the platform dashboard. We recommend always taking the address from the official dashboard rather than trusting addresses circulating on third-party pages.
Does data leave Türkiye?
Not according to the announcement. Prompts, responses and usage logs are processed in Türkiye and not transferred abroad.
How large is the context window?
The announcement states a context window of up to 1 million tokens. Check the dashboard to see whether this limit differs by model.
Is Aksiyon Soft an EVREN partner?
No. Aksiyon Soft has no partnership with SSB or EVREN. This article is industry news based on public sources.
Let’s talk about your project
If you want to connect a domestic LLM service to your existing application or build the control layer around it, write to us through our contact form and we will define the pilot scenario together in the first call.
Sources
Related news
Company and services
Kahta software coverage: enterprise solutions for tourism, farming and fisheries
Aksiyon Soft serves businesses in Adıyaman’s Kahta district — Nemrut tourism, farming and Atatürk Dam lake fisheries — remotely from Samsun with enterprise software and integration.
Company and services
Sinop software coverage: enterprise solutions for seafood, tourism and SMEs
From our HQ in neighbouring Samsun, we bring ERP, field apps, e-invoice and middleware integration to Sinop’s seafood processors, tourism operators and SMEs in Boyabat and Gerze.