d3

POST /v1/systemone model: d3

d3 is the 27B model of Decision 3.0, the decision models of vLLM Semantic Router, trained from Qwen3.8-27B. It reads text, JSON, images and videos. d3-flash is its 9B sibling with the same API: faster, and a little less accurate.

At a glance


Endpoint	`POST /v1/systemone`
Input	a `state` (text or any JSON), plus optional images and videos
Output	one answer per question, with a probability for every option
Question types	`choice`, `noul`, `score`
Input limit	32,768 tokens per question; longer requests are refused, not truncated
Streaming	❌
Billing	per input token, reported as `usage.input_tokens`; `output_tokens` is always `0`
Stability	`experimental`
License	Apache-2.0

Request

Field
`model`	required	`d3`
`state`	required	the situation to decide about: a string, or any JSON value
`questions`	required	an object mapping your own question names to questions, at least one
`images`	optional	an array of images as `data:` URL strings (PNG, JPEG or WebP)
`videos`	optional	an array of videos as `data:` URL strings (MP4, WebM, QuickTime or Matroska)

Every question sees all the images and videos of the request. The model card gives the budgets it reads them with: each image at up to 1.6 megapixels, and each video at 2 frames per second, at most 32 frames of up to 0.2 megapixels. The whole request body must stay under 5 MB.

The prompt around your input is fixed, so a top-level instructions field is refused with a 400.

Question types

`type`	`criteria`	Answer
`choice`	an object of options, `{"name": "what this option means", …}`	`choice`, `probabilities` per option, and `confidence`, the chosen option’s probability
`noul`	none; put the question in `instructions`	`noul`, the probability of yes; `probabilities` for `yes` and `no`; `confidence`
`score`	an ordered array of levels, lowest first	`score` (the expected level, counting from 0), `legend`, `probabilities` per level, and `confidence`, the most likely level’s probability

Every question carries instructions, the question itself in plain words.

Response

{
  "id": "decision-aa4a76300e0af8f5",
  "object": "structured_decision",
  "created": 1792166448,
  "model": "d3",
  "answers": {
    "team": {
      "type": "choice",
      "choice": "auth",
      "probabilities": { "auth": 0.9993, "billing": 0.0000, "other": 0.0006 },
      "confidence": 0.9993
    },
    "urgent": {
      "type": "noul",
      "noul": 0.9967,
      "probabilities": { "yes": 0.9967, "no": 0.0033 },
      "confidence": 0.9967
    },
    "severity": {
      "type": "score",
      "score": 1.8641,
      "legend": { "0": "Low", "1": "Medium", "2": "High" },
      "probabilities": { "0": 0.0034, "1": 0.1292, "2": 0.8674 },
      "confidence": 0.8674
    }
  },
  "usage": { "input_tokens": 352, "output_tokens": 0 },
  "diagnostics": {
    "team": { "label_mass": 1.0, "argmax_is_label": true },
    "urgent": { "label_mass": 1.0, "argmax_is_label": true },
    "severity": { "label_mass": 1.0, "argmax_is_label": true }
  }
}

diagnostics reports, per question, the share of probability that fell on the question’s answer labels (label_mass) and whether the single most likely code was one of them (argmax_is_label).

Use the probabilities rather than only the top answer: a noul of 0.55 and one of 0.99 are both “yes”, but only one of them is safe to act on automatically.

Errors

Errors use the usual {"error": { … }} shape:

Status	Example `message`
400	`unknown question type 'bogus'; supported: ['choice', 'noul', 'score']`
400	`question 'q': the prompt has 33356 tokens, more than max_model_len=32768; nothing is truncated`
400	`this model reads a fixed prompt, so it takes no instructions`
401	the API key is missing or wrong

Example

curl https://developer.amd.com.cn/radeon/api/v1/systemone \
  -H "Authorization: Bearer $RADEON_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "d3",
    "state": {"ticket": "Customer cannot log in after resetting the password; error 403 on every attempt."},
    "questions": {
      "team": {"type": "choice", "instructions": "Which team should handle this?",
               "criteria": {"auth": "Identity and login", "billing": "Payments and invoices", "other": "Anything else"}},
      "urgent": {"type": "noul", "instructions": "Is the customer blocked from using the product?"},
      "severity": {"type": "score", "instructions": "How severe is this?", "criteria": ["Low", "Medium", "High"]}
    }
  }'

With a video:

import base64, requests

clip = base64.b64encode(open("clip.mp4", "rb").read()).decode()
r = requests.post(
    "https://developer.amd.com.cn/radeon/api/v1/systemone",
    headers={"Authorization": f"Bearer {API_KEY}"},
    json={
        "model": "d3",
        "state": "Watch the clip.",
        "videos": [f"data:video/mp4;base64,{clip}"],
        "questions": {
            "direction": {"type": "choice", "instructions": "Which way does the ball move?",
                          "criteria": {"left_to_right": "Left to right", "right_to_left": "Right to left"}},
        },
    },
    timeout=120,
)
print(r.json()["answers"]["direction"])