LLMTR Trendyol Asure 12B

Trendyol Asure 12B is a 12B chat model produced by Trendyol and hosted by us in Turkey. It is called through /v1/chat/completions with the canonical model id llmtr/trendyol-asure-12b. It fits Turkish chat, RAG chatbots, customer-support flows, summarization, and e-commerce text work.

Property	Value
Model id	llmtr/trendyol-asure-12b
Context window	40,960 tokens
Price (input / output / cache read)	$0.10 / $0.50 / $0.025 per 1M tokens
Image input	Yes (base64 data URL only)
Prompt cache	Yes
Tool calling	No
Reasoning selector	No
Audio / video input	No
Image generation	No

curl "$LLMTR_BASE_URL/v1/chat/completions" \
  -H "Authorization: Bearer llmtr-your_key" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "llmtr/trendyol-asure-12b",
    "messages": [
      { "role": "user", "content": "Summarize this product description in two sentences." }
    ],
    "max_tokens": 512
  }'

Image input

The model can read images, but it accepts them only as base64 data URLs. If you send a remote https:// address, the gateway rejects the request with 400 unsupported_input without dialing the upstream:

{
  "error": {
    "message": "Model accepts image input only as base64 image data URLs",
    "type": "unsupported_input"
  }
}

Correct usage:

{
  "model": "llmtr/trendyol-asure-12b",
  "messages": [
    {
      "role": "user",
      "content": [
        { "type": "text", "text": "What is in this image?" },
        {
          "type": "image_url",
          "image_url": { "url": "data:image/png;base64,iVBORw0KGgo..." }
        }
      ]
    }
  ]
}

Image tokens are billed as ordinary input tokens; there is no separate image rate.

Prompt cache

Repeated prompt prefixes (a long system prompt, a fixed RAG context, an agent loop) are served from the upstream cache. Cached tokens are reported in prompt_tokens_details.cached_tokens and billed at the cache read rate instead of the input rate:

"usage": {
  "prompt_tokens": 1595,
  "completion_tokens": 8,
  "total_tokens": 1603,
  "prompt_tokens_details": { "cached_tokens": 1582 }
}

Here 1582 tokens are billed at $0.025/1M and the remaining 13 at $0.10/1M. Requests without a cache hit behave exactly as before.

Tool calling is not supported

Sending tools to this model does not return structured tool calls, and the request is not rejected either: it answers 200, omits tool_calls, and replies in prose. The model may fabricate the information the tool was supposed to provide.

Measured example: with a get_weather tool defined and asked "What is the weather in Ankara?", the model answered "Ankara is cloudy and rainy, around 5°C" instead of requesting the tool. That reading comes from no real source.

A client that only checks the HTTP status would treat this fabrication as a valid answer. For work that needs tool calling, use one of the supported models listed on the Tool calling page; if data must be processed in Turkey, llmtr/qwen3-5-4b supports tool calling.

Migrating from Trendyol 7B

llmtr/trendyol-7b is being replaced by this model.

• Until 2026-09-15: calls asking for llmtr/trendyol-7b automatically run on llmtr/trendyol-asure-12b. The list price is identical, so your bill does not change.
• From 2026-09-15: the old identifier returns 410 and the response body names the successor.

Update the model id in your integration to llmtr/trendyol-asure-12b today. Differences between the two models:

	Trendyol 7B	Trendyol Asure 12B
Context window	32,768 tokens	40,960 tokens
Image input	No	Yes (base64 data URL)
Prompt cache	No	Yes
Price (input / output)	$0.10 / $0.50	$0.10 / $0.50