# Decision 3.0 (slim)

> Decision 3.0 is a family of open-weight multimodal decision models from the vLLM Semantic Router team, 0.6B to 27B, under Apache-2.0. They answer choice, yes-or-no and score questions about text, images and video with a probability per option.

- Full: https://www.anchorterminal.com/tools/decision-3.md (~6,400 tokens) · this version ~1,430 tokens · JSON https://www.anchorterminal.com/tools/decision-3.json · canonical https://www.anchorterminal.com/tools/decision-3
- Index: https://www.anchorterminal.com/llms.txt · API: https://www.anchorterminal.com/api/v1/index.json · Updated: 2026-10-12

**B · 63.1/100 · rank #413 of 965 · #8 in Decision models · not agent-ready · confidence medium**


Assessment: Open Apache-2.0 decision models that read images and video as well as text and answer several typed questions in one call, with a probability per option and tagged Hub releases. They are two days old, need a GPU and `trust_remote_code`, drop Vela's set and span types, and publish no calibration figures.

## Facts

- Kind: Model API · vendor: vLLM Semantic Router Team · category: Decision models · legal entity: not named · provenance 27/100
- Local only (HTTP): pypi `vllm-sr`
- Auth: None · pricing: Free · x402: no · licence: Apache-2.0 for the weights and the bundled runtime, built on Apache-2.0 Qwen3.5 and Qwen3.8 bases
- Probe metrics: not measured yet (probes haven't run)
- Models: d3 (26.09B with a 0.46B vision encoder, on Qwen3.8-27B), d3-flash (8.39B, Qwen3.5-9B), d3-mini (4.54B, Qwen3.5-4B), d3-nano (2.21B, Qwen3.5-2B), d3-lite (0.85B, Qwen3.5-0.8B) and d3-edge (0.59B, a pruned d3-lite, outside the collection)
- Question types: choice (up to 255 options), noul (yes or no) and score (ordered levels), any mix in one request, one forward pass per question, a probability for every option and no generated text
- Inputs: Text or JSON, plus any number of images (PNG, JPEG or WebP, read at up to 1.6 MP) and videos (MP4, WebM, MOV or MKV, 2 frames a second, at most 32 frames and 16,384 video tokens a request)
- Serving: `AutoModel.from_pretrained(..., trust_remote_code=True)` and `model.system_one(...)`, the bundled `d3_server.py` on `POST /v1/systemone`, or the router's model runtime on `POST /v1/decisions` (Engine mode, development channel)
- Server limits: Data URLs only, each image at most 8,000,000 bytes and 16,000,000 pixels, each video at most 32,000,000 bytes and 300 seconds. Requests are served one at a time
- Errors: Bundled server 401, 422 (malformed body, unknown fields, invalid questions or over the input limit) and 503 while loading. Model runtime `{error: {code, message}}`
- Licence: Apache-2.0 with a NOTICE naming the Qwen base, which is also Apache-2.0
- Hosted option: None from the authors

- Scores: Reliability 58, Performance pending, Schema & documentation 70, Agent ergonomics 72, Security & auth 56, Payments & pricing 60, Task success pending, Maintenance & community 82, Transparency & trust 45 · total over the 7 assessed categories
- Why: Reliability, Scored on the local-package checklist, since Decision 3.0 is a model the owner runs. · Schema & documentation, Read for a model the owner serves. · Agent ergonomics, Read as an API an agent calls for a decision, as with Vela, Decider and Kev. · Security & auth, Read as software the owner runs. · Payments & pricing, Free Apache-2.0 software with nothing to buy, so 20 + 20 + 20 for pricing, free use and no sign-up. · Maintenance & community, Read for an open-weight model. · Transparency & trust, Apache-2.0 for the weights and runtime, on Apache-2.0 Qwen3.5 and Qwen3.8 bases, with a NOTICE in each repository.
- Sources: 14, open questions: 7, both in the full twin
- Capabilities: inference.decision
- JSON: https://www.anchorterminal.com/api/v1/tools/decision-3.json
- Verify (for the vendor): the badge `https://www.anchorterminal.com/badges/decision-3.svg` or a link to https://www.anchorterminal.com/tools/decision-3 from a page on vllm-sr.ai or one of its subdomains, or the README of github.com/vllm-project/semantic-router, then `POST https://www.anchorterminal.com/api/v1/verify` `{"slug", "url"}` or `verify_listing` at /mcp; re-checked weekly, no effect on the grade. Snippets in the full twin.

## Before you call it

1. Pin a tag or commit with `revision=` when loading from the Hub. Each model had five tagged releases on 10 and 11 October
2. Send images and videos to `d3_server.py` as base64 data URLs only. The server refuses paths and http URLs with 422
3. Set `DECISION_API_KEY` before binding the bundled server beyond 127.0.0.1. It is open otherwise
4. Turn on `permutation_average=True` when routing must not depend on option order. It is off by default
5. Expect one request at a time from the bundled server. Use the router's model runtime for queued serving

## Connect

```bash
pip install "transformers==5.17.0" torch torchvision pillow opencv-python-headless safetensors accelerate
pip install flash-linear-attention  # optional: fast GPU kernels for the linear-attention layers
```

```bash
pip install fastapi uvicorn
python d3_server.py --model vllm-sr/d3-mini --device cuda:0 --port 8000
```

## Similar tools

| Tool | Grade | Score | Shared capabilities | Slim |
| --- | --- | --- | --- | --- |
| OpenAI Decisions API | BB | 71.5 | inference.decision | https://www.anchorterminal.com/tools/openai-decisions-api.min.md |
| Drex 1.5 | B | 69.7 | inference.decision | https://www.anchorterminal.com/tools/nace-drex.min.md |
| Decider | B | 69.5 | inference.decision | https://www.anchorterminal.com/tools/decider.min.md |
| Laya | B | 69.2 | inference.decision | https://www.anchorterminal.com/tools/convai-laya.min.md |
| Kev | B | 67.4 | inference.decision | https://www.anchorterminal.com/tools/jaredpalmer-kev.min.md |


## Panel reviews (0, desk reviews from public material, no calls made)