Engy

Engy is an inference platform that serves open-weight models over an OpenAI-compatible API. The LLMTR catalog carries two Engy models; both are called on /v1/chat/completions with the engy/ prefix, exactly as the catalog lists them.

Model	Context	Input / Cache read / Output ($/1M)
engy/kimi-k3-engy	1,113,088	1.95 / 0.195 / 9.75
engy/deepseek-v4.1-flash-1017-engy	1,048,576	0.03 / 0.006 / 0.12

Repeated prompt prefixes can be served from a cache; those tokens bill at the cache-read rate in the table. The other providers serving the same model are listed on each model page, in the menu next to the provider name.

Chat

curl "$LLMTR_BASE_URL/v1/chat/completions" \
  -H "Authorization: Bearer llmtr-your_key" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "engy/kimi-k3-engy",
    "messages": [
      { "role": "system", "content": "Answer briefly and clearly." },
      { "role": "user", "content": "Summarise the obligations in this contract clause in three points." }
    ],
    "max_tokens": 4096
  }'

Both models support tool/function calling and streaming.

Reasoning

Both models reason by default, and those tokens bill as output. To turn reasoning off, append :fast to the model id (engy/deepseek-v4.1-flash-1017-engy:fast) or send "reasoning": false in the request. With it off, no reasoning tokens are produced or billed.

reasoning_effort is not supported on these models; a request that sends it is refused with 400 invalid_request.

If an answer returns with finish_reason: "length", raise max_tokens and retry.

Image input

Both models accept images. Images go in an image_url part as a base64 data URL. An https:// image URL is not accepted; LLMTR refuses the request with 400 unsupported_input before it reaches the provider, and it is not billed.

IMAGE_B64=$(base64 -w0 invoice.png)
curl "$LLMTR_BASE_URL/v1/chat/completions" \
  -H "Authorization: Bearer llmtr-your_key" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "engy/deepseek-v4.1-flash-1017-engy",
    "messages": [
      {
        "role": "user",
        "content": [
          { "type": "text", "text": "What is the total amount on this invoice?" },
          { "type": "image_url", "image_url": { "url": "data:image/png;base64,'"$IMAGE_B64"'" } }
        ]
      }
    ],
    "max_tokens": 2048
  }'

Tool calling

On engy/deepseek-v4.1-flash-1017-engy, tool_choice honours auto, required and a named function.

On engy/kimi-k3-engy, required and a named function are honoured on most requests, but this is not guaranteed: the answer can come back without a tool call or with a different tool. On this model, check the tool_calls field and the name of the function called in the response, and retry the request if the call you expect is missing.

On engy/deepseek-v4.1-flash-1017-engy, tool_choice: "none" with tools attached is refused with 400 unsupported_input; to turn tool use off on that model, leave the tools field out entirely.