INTRODUCING

Decision 3.0

d3 27B

Towards Open Multimodal Foundation Decision Models

ONE CALL · MANY DECISIONS

Every input.
One call.

Give d3 text or JSON with images and videos, and ask the questions you need answered. It answers them all in one call and returns a probability for every option, without generating text.

Choice: Pick an option
Yes / No: A probability of yes
Score: A level on your scale

Text or JSON Any context, shared by every question. Nothing is truncated.
Images Several per request: PNG, JPEG or WebP, each read at up to 1.6 MP.
Video MP4, WebM, MOV or MKV, read at 2 frames per second, up to 32 frames.

THE STUDIO

Your input. Your decisions.

Pick an example, or drop your own image or video, and ask Choice, Yes / No and Score questions about it.

Text, images and video in one request.

Model

d327B v3.1.0Checking

Auto-run

Try an example

01Input

Example

Context

Every question sees the context and all media

Images are read at up to 1.6 MP; videos at 2 frames per second, up to 32 frames. Nothing is truncated.

02Decisions

—

Ready when you are

Meet d3.

d3 is the 27B multimodal foundation decision model of Decision 3.0. Give it text or JSON with images and videos and the questions you need answered; it returns a probability for every option.

Parameters: 26.09B, including the 0.46B vision encoder
Inputs: Text or JSON, images and videos (several per request)
Decision types: Choice · Yes / No · Score
Release: v3.1.0
Serving here: AMD Instinct MI300X
License: Apache-2.0

Quickstart

from transformers import AutoModel

model = AutoModel.from_pretrained(
    "vllm-sr/d3", revision="v3.1.0", trust_remote_code=True
)
result = model.system_one(
    state="The blender arrived cracked; the receipt is attached.",
    images=["receipt.png"],    # paths, URLs, PIL images or data URLs
    videos=["unboxing.mp4"],   # read at 2 frames per second
    questions={
        "route": {
            "type": "choice",
            "instructions": "Which team should handle this?",
            "criteria": {"returns": None, "billing": None},
        },
        "on_receipt": {
            "type": "noul",
            "instructions": "Does the receipt list the blender?",
        },
        "urgency": {
            "type": "score",
            "instructions": "How urgent is it?",
            "criteria": ["Routine", "Soon", "Today"],
        },
    },
)
print(result["answers"])