LLMTR Muse Glimmer 30B (Turkey)

Muse Glimmer 30B is our Turkey-hosted deployment of Meta's Apache 2.0 open-weight model. It is called through /v1/chat/completions with the canonical model id llmtr/muse-glimmer-30b-tr. It fits agent loops, tool-using workflows, and reasoning tasks whose data must be processed in Turkey.

Property	Value
Model id	llmtr/muse-glimmer-30b-tr
Context window	131,072 tokens (128K)
Price (input / output / cache read)	$2.00 / $5.00 / $0.50 per 1M tokens
Image input	Yes (base64 data URL only)
Tool calling	Yes
Prompt cache	Yes
Reasoning	Yes, always on — cannot be disabled
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/muse-glimmer-30b-tr",
    "messages": [
      { "role": "user", "content": "Find the bug in this snippet." }
    ],
    "max_tokens": 2048
  }'

Reasoning is always on

The model reasons step by step before answering, and that reasoning is returned to you in message.reasoning_content. This cannot be turned off; there is no :fast selector.

Reasoning tokens are spent from your max_tokens budget and billed as output. Even when you expect a short answer, budget for the reasoning pass; with a very small max_tokens the budget is consumed by reasoning and content can come back empty.

Image input

The model reads images, but it accepts them only as base64 data URLs. A remote https:// address returns 400 unsupported_input.

{
  "model": "llmtr/muse-glimmer-30b-tr",
  "messages": [
    {
      "role": "user",
      "content": [
        { "type": "text", "text": "What is in this image?" },
        {
          "type": "image_url",
          "image_url": { "url": "data:image/png;base64,iVBORw0KGgo..." }
        }
      ]
    }
  ],
  "max_tokens": 2048
}

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

Tool calling

The model supports the standard OpenAI tools schema and returns real tool_calls:

{
  "choices": [
    {
      "message": {
        "role": "assistant",
        "content": "",
        "tool_calls": [
          {
            "id": "call_1",
            "type": "function",
            "function": { "name": "get_weather", "arguments": "{\"city\":\"Ankara\"}" }
          }
        ]
      },
      "finish_reason": "tool_calls"
    }
  ]
}

In multi-step flows, append the assistant message to the history with both its tool_calls and its reasoning_content.

Prompt cache

Repeated prompt prefixes are served from 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": 1096,
  "completion_tokens": 12,
  "prompt_tokens_details": { "cached_tokens": 1080 }
}

Here 1080 tokens are billed at $0.50/1M and the remaining 16 at $2.00/1M. A cache hit also cuts latency substantially.

Repeating an identical request

A byte-identical request repeated within a short window may return the previous answer verbatim. The response then arrives almost instantly but the content does not change — even with a high temperature.

If you need a different answer on each call, vary something in the request body. Replayed responses are billed at the normal rate.

Model identity

The catalog carries two rows from the same model family. The row on this page is hosted in Turkey and is called with the id llmtr/muse-glimmer-30b-tr. For the other row, see Meta Muse Glimmer.

The two ids are not interchangeable; be explicit about which one your integration uses.