EVREN

EVREN is the LLM inference service opened by the SAYZEK programme of Turkey's Presidency of Defence Industries, hosted in Turkey. The EVREN rows on LLMTR run on your own EVREN API key: the request is made with your credential, LLMTR only routes it, and LLMTR charges nothing for these requests.

Model	Context	Price ($/1M)
evren/gemma-4-31b	256,000	0
evren/deepseek-v4-flash-tr	1,000,000	0
evren/qwen3.8-flash-next	256,000	0
evren/glm-5.3-fp8	512,000	0
evren/qwen3-vl-30b	256,000	0
evren/mimo-v2.6-pro-tr	1,048,576	0
evren/deepseek-v4.1-flash-tr	1,048,576	0

Register your key first

Unlike every other model on LLMTR, these rows do not answer until you have stored an EVREN key.

1. Create an API key in the EVREN console.
2. Paste it into Settings → Your own provider keys on LLMTR.
3. When you save the key, LLMTR asks EVREN whether its terms of use have been accepted for it. If they have not, you are asked "Do you accept the EVREN terms of use?"; the terms can be read on the EVREN site. Once you press I accept, the key is ready to use. The card always shows where the terms stand, with an Accept button when they are not accepted.
4. Call the model as usual. You do not send the EVREN key in your request; only your LLMTR key travels.

Once saved, a key is never shown again — only its last four characters. If you lose it, create a new one in the EVREN console and replace it in Settings.

curl "$LLMTR_BASE_URL/v1/chat/completions" \
  -H "Authorization: Bearer llmtr-your_key" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "evren/glm-5.3-fp8",
    "messages": [
      { "role": "user", "content": "Summarise the termination clauses in this contract." }
    ],
    "max_tokens": 4096
  }'

MiMo V2.6 Pro

evren/mimo-v2.6-pro-tr is Xiaomi's MiMo V2.6 Pro served by EVREN. It offers tool calling, JSON mode, schema-conforming structured output and the option to turn reasoning off.

• Reasoning is on by default. To turn it off, send none or minimal in the reasoning_effort field or as a suffix on the model id. To keep it on, leave the field out; other levels are refused with 400. reasoning: { "enabled": false } and include_reasoning: false do not turn it off, and reasoning: false is refused with 400; a nested reasoning: { "effort": "none" } is handled like reasoning_effort.
• Reasoning tokens are drawn from the output budget (max_tokens or max_completion_tokens). Keep it generous while reasoning is on; if you get an empty content, raise the output budget you sent or retry with reasoning off.
• Tool calling works with auto, required and a named choice for tool_choice; a response that carries tool calls comes back with finish_reason: "tool_calls". tool_choice: "none" is refused with 400 on a request that carries tools; to turn tool use off, leave the tools field out.
• Image input is accepted; send the image as a base64 data URL. An image sent as an http(s) address is refused with 400. Audio and video input are not offered on this row.
• The 1,048,576-token window is shared by input and output. There is no separate output ceiling; a long prompt leaves less room for the output. If you send neither max_tokens nor max_completion_tokens, the answer stops at 16,384 tokens; set one of them explicitly for a longer output. If you send both, max_completion_tokens applies.

curl "$LLMTR_BASE_URL/v1/chat/completions" \
  -H "Authorization: Bearer llmtr-your_key" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "evren/mimo-v2.6-pro-tr:none",
    "messages": [
      { "role": "user", "content": "What is the weather in Ankara?" }
    ],
    "tools": [
      {
        "type": "function",
        "function": {
          "name": "get_weather",
          "parameters": {
            "type": "object",
            "properties": { "city": { "type": "string" } },
            "required": ["city"]
          }
        }
      }
    ],
    "max_tokens": 1024
  }'

DeepSeek V4.1 Flash

evren/deepseek-v4.1-flash-tr is DeepSeek V4.1 Flash served by EVREN. It offers tool calling, JSON mode, schema-conforming structured output and the option to turn reasoning off.

• Reasoning is on by default. To turn it off, send none in the reasoning_effort field or as a suffix on the model id. To keep it on, leave the field out; other levels are refused with 400. reasoning: { "enabled": false } and include_reasoning: false do not turn it off, and reasoning: false is refused with 400; a nested reasoning: { "effort": "none" } is handled like reasoning_effort.
• Reasoning tokens are drawn from the output budget (max_tokens or max_completion_tokens). Keep it generous while reasoning is on; if you get an empty content, raise the output budget you sent or retry with reasoning off.
• Tool calling works with auto, required and a named choice for tool_choice; a response that carries tool calls comes back with finish_reason: "tool_calls", and one response can call several tools. tool_choice: "none" is refused with 400 on a request that carries tools; to turn tool use off, leave the tools field out.
• Image input is accepted; send the image as a base64 data URL. An image sent as an http(s) address is refused with 400.
• The 1,048,576-token window is shared by input and output. There is no separate output ceiling. If you send neither max_tokens nor max_completion_tokens, the answer stops at 16,384 tokens; set one of them explicitly for a longer output. If you send both, max_completion_tokens applies.

curl "$LLMTR_BASE_URL/v1/chat/completions"   -H "Authorization: Bearer llmtr-your_key"   -H "Content-Type: application/json"   -d '{
    "model": "evren/deepseek-v4.1-flash-tr",
    "messages": [
      { "role": "user", "content": "Find the bug in this function and write the corrected version." }
    ],
    "max_tokens": 8192
  }'

Keeping your key safe

A stored key is held under strong encryption and is used only on your behalf: to send your own requests, to ask whether EVREN's terms of use are accepted for this key, and to accept them once you press I accept. It is never shown again, not even to you — only its last four characters. The moment you remove it, it can no longer be used.

Error responses

On EVREN rows the provider's answer is relayed to you directly. The key and the account are yours, so you are the only person who can act on either.

Situation	HTTP	error.type
No key stored	403	provider_credential_required
EVREN rejected your key	403	provider_credential_rejected
EVREN's terms of use have not been accepted for this key	403	provider_terms_required
Your EVREN account did not serve the request	403	provider_account_limited
EVREN rate limited the request	429	provider_rate_limited
EVREN refused the request for another reason (4xx)	the provider's code	provider_upstream_error
EVREN failed on its side (5xx)	502	provider_upstream_error

When you get provider_terms_required, accept the terms on the Settings → Your own provider keys card (LLMTR Chat sends you there too), then send your request again. The terms cannot be accepted through an API request, only in the dashboard.

error.details.upstreamStatus always carries the original status EVREN returned, even where the HTTP status is 502. The distinction matters because clients decide retries from the HTTP status: 4xx codes are preserved as they are, 5xx responses are collapsed into 502. The provider's error text is not relayed; only the status and which provider answered.

Things worth knowing

• Streaming on these rows is not real streaming: the answer is delivered in chunks after it completes.
• Tool calling, JSON mode and reasoning controls are not offered on the rows other than evren/mimo-v2.6-pro-tr and evren/deepseek-v4.1-flash-tr.
• For the other rows, context windows and parameter counts are taken from EVREN's published information.
• For the other rows, no separate output ceiling is published. The limit shown on the model card is the context window; if EVREN's own output limit is lower, it refuses the request and its answer is relayed to you. If you expect long outputs, raise max_tokens gradually.
• The service being free is EVREN's decision. It is free until 1 November 2026; LLMTR adds no charge of its own. The platform margin on credit top-ups is unaffected.