Model comparison · 2026-09-20
Turkey-hosted LLM models: what sits beside EVREN in the catalog
The LLMTR catalog holds twelve live rows marked as hosted in Turkey; five are EVREN, seven are not. The full list with context, price, modality and tool-calling columns, and how the two groups are billed.
Short answer
EVREN is not the only Turkey-hosted option. The LLMTR catalog holds twelve live rows marked as hosted in Turkey: five are EVREN, seven are LLMTR's own models.
Both groups are called from the same endpoint and authenticate with the same LLMTR key. They differ in billing: EVREN rows run on your own EVREN key and LLMTR does not charge for them, while LLMTR rows draw down your balance at the catalog price.
Turkey rows in the catalog
The table below is derived from the catalog itself; it is not a hand-picked list. When a new Turkey row is added, or a row retires, this table has to change with it.
The price column shows the rule in effect today, in USD per 1M tokens for input and output. The embedding row has no output token price.
| Model id | Name | Context | Input / Output | Input | Tool calling |
|---|---|---|---|---|---|
| evren/deepseek-v4-flash-tr | DeepSeek V4 Flash | 1,000,000 | $0 / $0 | Text | No |
| evren/glm-5.3-fp8 | GLM-5.3 | 512,000 | $0 / $0 | Text | No |
| evren/gemma-4-31b | Gemma 4 31B | 256,000 | $0 / $0 | Text, image | No |
| evren/qwen3.8-flash-next | Qwen3.8 Flash Next | 256,000 | $0 / $0 | Text, image, video | No |
| evren/qwen3-vl-30b | Qwen3 VL 30B | 256,000 | $0 / $0 | Text, image, video | No |
| llmtr/qwen3-6-35b | Qwen 3.6 35B-A3B | 262,144 | $5.00 / $5.00 | Text | Yes |
| llmtr/gemma-4 | Gemma 4 | 131,072 | $2.00 / $5.00 | Text, image | Yes |
| llmtr/muse-glimmer-30b-tr | Muse Glimmer 30B (Türkiye) | 131,072 | $2.00 / $5.00 | Text, image | Yes |
| llmtr/qwen3-5-4b | Qwen 3.5 4B | 131,072 | $1.00 / $3.00 | Text | Yes |
| llmtr/trendyol-asure-12b | Trendyol Asure 12B | 40,960 | $0.10 / $0.50 | Text, image | No |
| llmtr/magibu-11b-v8 | Magibu 11B v8 | 8,192 | $0.10 / $0.50 | Text | No |
| llmtr/embeddinggemma-300m | EmbeddingGemma 300M | 2,048 | $0.10 | Text, vector output | No |
Two groups, two billing shapes
The catalog price of EVREN rows is zero, and this is not a promotion on our side: the request goes with your EVREN key, from your EVREN account. The service itself is free until 1 November 2026 according to its own announcement; what happens after that is EVREN's decision and does not change our rates.
LLMTR rows are billed at the catalog price and draw down your credit balance. The 8 percent platform share added when you load credit is not added to model rates; it applies only to the loaded amount.
The practical consequence: EVREN rows need an EVREN account and key first, LLMTR rows need only a balance.
Differences in context and capability
Context windows diverge sharply between the two groups. The declared windows on EVREN rows run from 256,000 to 1,000,000 tokens; the windows on LLMTR rows run from 2,048 to 262,144. If you process long documents, that difference alone can decide the choice.
The tool-calling column states whether the row offers tool calling today. Sending a tool definition to a row that does not may not return an error; the model can answer in prose instead of calling the tool. If you are building an agent loop, read this column before you call.
The input column only states which types the model accepts. How to supply image input on rows that accept it is documented on the model card.
Which one to pick
If you want a long context at zero cost and can open an EVREN account, the EVREN rows are strong. Their windows are among the widest in the catalog and there is no charge from us.
If you are building an agent loop that needs tool calling, the rows that offer it today are on the LLMTR side: four LLMTR rows carry the capability.
If you want a small, cheap model specialised for Turkish, the Trendyol Asure and Magibu rows start at ten cents per 1M tokens. For embedding work, a Turkey-hosted vector model sits in the same catalog.
Why through one API?
All twelve rows in this table are called with one key, from one endpoint. Nothing in the request body changes except the model id.
You can connect to EVREN directly; what connecting through LLMTR changes is that the same code also reaches other Turkey-hosted models and global models. When one job has to run on a Turkey-hosted model and another on a global one, you do not write a second integration.
If the decision depends on where data is processed, every model card states where the model runs and what the provider may record. Make the choice from the card, not by guessing from the model name.
Frequently asked questions
Is EVREN the only Turkey-hosted LLM API?
No. The LLMTR catalog holds twelve live rows marked as hosted in Turkey; five are EVREN rows and seven are LLMTR's own models. You can connect to EVREN directly; what connecting through LLMTR changes is that the same code also reaches other Turkey-hosted models and global models.
Are all of these models called with the same key?
For authentication, yes: every request carries only your LLMTR key. EVREN rows additionally require an EVREN key registered once; that key does not travel in your request, it stays registered in Settings.
Why are EVREN rows priced at zero?
Because the request goes with your EVREN key, from your EVREN account, and LLMTR does not charge for it. That the service itself is free is EVREN's decision, and according to its announcement it holds until 1 November 2026.
Why are retired Turkey models missing from this list?
The list shows live rows. A retired row is not deleted from the catalog and keeps its own page, but it accepts no new requests, so it has no place in a selection table; a table that points at a retired row is worse than no table.
What changes when I move to a global model?
The model field in the request body. The endpoint, the authorization header and the body schema stay the same. Global rows draw down your balance and carry no Turkey-hosting claim.