Models

GET /v1/models

Returns the model behind this deployment endpoint, OpenAI-style. Because each endpoint serves exactly one model, the list always has one entry — useful for SDKs and tools that discover models via this route, and for grabbing the exact id to pass as model.

curl https://<your-deployment>.inference.meshive.ai/v1/models \
  -H "Authorization: Bearer $MESHIVE_API_KEY"

{
  "object": "list",
  "data": [
    {
      "id": "qwen2.5-7b-instruct-a1b2c3",
      "display_name": "Qwen/Qwen2.5-7B-Instruct",
      "object": "model",
      "created": 1780000000,
      "owned_by": "meshive",
      "permission": []
    }
  ]
}

id is the value to use in model parameters (when you choose to set them).

supported_parameters

Image and video deployments include an extra supported_parameters object (a Meshive extension — not part of the OpenAI schema). It lists, per generation task, which request parameters the model actually applies, so you can tell up front whether e.g. negative_prompt or guidance_scale will have any effect. Parameters not listed are accepted but ignored.

{
  "id": "qwen-image-a1b2c3",
  "object": "model",
  "supported_parameters": {
    "txt2img": {
      "parameters": ["guidance_scale", "n", "negative_prompt", "num_inference_steps", "prompt", "seed", "size"],
      "num_inference_steps": { "min": 1, "max": 50, "default": 30 }
    }
  }
}

• Task keys are txt2img / img2img for image deployments, t2v / i2v for video.
• When num_inference_steps is supported, its accepted range and default are included — out-of-range values are clamped.
• LLM and embedding deployments don’t carry this field.