ollama
- source id
- ollama
- execution mode
- local
- parameter count is approximate
- false
- quantization level
- BF16
- context window
- 125000
- context window is approximate
- true
- artifact size bytes
- 4351874238
- updated at
- 2026-11-16T08:10:06.079865+00:00
`d1-omni` is a 600M decision model from Liquid AI, built on LFM2.5-Encoder-350M. Give it a state (text or JSON, with images or a short voice clip) and a set of typed questions, and it returns a calib…
ollama/d1-omni
`d1-omni` is a 600M decision model from Liquid AI, built on LFM2.5-Encoder-350M. Give it a state (text or JSON, with images or a short voice clip) and a set of typed questions, and it returns a calibrated probability for every answer in a single forward pass.
It works through Ollama's [`/v1/systemone`](https://docs.ollama.com/capabilities/decision) endpoint and `ollama run --ask`.
## Highlights
- **Edge-sized**: 587M parameters, a 0.85 GB download. The same trunk handles every modality, with small vision (94M) and audio (112M) encoders. - **Images**: one or more images per state; large frames are tiled. - **Speech**: up to 30 seconds of English speech per request. - **One forward pass**: every option of every question is scored together, with no output tokens.
For the most accurate decisions over text and images, use [d1](https://ollama.com/library/d1), Liquid AI's 3B model. Use d1-omni when size matters or you need audio.
## Usage
``` ollama run d1-omni --ask "Is this support request urgent?" < ticket.txt ```
``` curl http://localhost:11434/v1/systemone -d '{ "model": "d1-omni", "state": "Our app crashes on launch for every user since this morning'\''s update. Please fix it now.", "questions": { "urgent": {"type": "noul", "instructions": "Is this support request urgent?"}, "team": { "type": "choice", "instructions": "Which team should handle this request?", "criteria": { "billing": "Payments, invoices, and refunds", "technical": "Outages, errors, and configuration", "sales": "Plans and upgrades" } } } }' ```
d1-omni answers `urgent` with 0.998 and picks `technical` with probability 0.995.
### Speech
Put one base64-encoded WAV clip in `images`, and send audio or images, not both. Clips are cut at 30 seconds. The audio training data was English requests from a speaker to an assistant, so questions about what kind of utterance it is, its topic, and what the speaker wants work best.
``` curl http://localhost:11434/v1/systemone -d '{ "model": "d1-omni", "state": "Voice message from a customer.", "images": ["<base64-encoded WAV>"], "questions": { "intent": { "type": "choice", "instructions": "What does the speaker want?", "criteria": {"timer": "Set a timer or alarm", "music": "Play music", "weather": "Hear the weather", "other": null} } } }' ```
### Images
The model was trained with up to 896 tokens of text alongside images, so keep the state and questions short when you send images.
## Benchmarks
Accuracy reported by Liquid AI on public benchmarks framed as decisions. See the [model card](https://huggingface.co/LiquidAI/d1-omni-600M) for details.
| Benchmark | d1-omni | d1 | |---|---:|---:| | Civil Comments | **95.8** | 93.0 | | MASSIVE intent | 86.1 | **87.3** | | PAWS-X | **79.5** | 76.9 | | BoolQ | 77.7 | **86.7** | | SQuAD 2.0 | 74.0 | **85.3** | | **Mean (7 benchmarks)** | 78.4 | **82.9** |
On the harder Decision Index 0.2.1, Liquid AI reports 15.95 for d1-omni and 48.57 for d1.
## License
[LFM Open License v1.0](https://huggingface.co/LiquidAI/d1-omni-600M/blob/main/LICENSE)
1–8 / 8
1–2 / 2
1–1 / 1
These charts contain model-linked results. The charts above also include unlinked source profiles. Compare scores within each evaluation. Model quality, usage, transparency and system performance measure different things.