Open d1 (d1-3B, d1-omni-600M)
Liquid AI's open-weight decision models that return a calibrated yes/no/choice/score answer in a single forward pass instead of generating tokens.
Overview
Liquid AI released Open d1 on October 7, 2026, two open-weight 'decision' models under the LFM Open License v1.0: d1-3B (3.1B parameters, text and image input) and d1-omni-600M (600M parameters, an early research release accepting text plus image or audio). Unlike generative LLMs, both emit zero output tokens, instead returning a calibrated probability for a typed question (yes/no, choice, or score) in one forward pass, with d1-3B answering in roughly 8ms on an RTX 4090. On Liquid AI's internal Decision Index v0.2.1 benchmark, d1-3B scored 48.57, beating the 12x-larger Decider-35B-A3B's 47.11. On October 15, 2026 Liquid AI added d1-ultra, a closed hosted decision API for text, images, or both that returns probabilities for yes/no, label-selection, and scoring questions without generating output tokens, billed at $0.042 per million input tokens with each question counted as a separate prompt; Vercel and OpenRouter availability was listed as forthcoming.
This is a sourced directory entry. Features and capabilities are reported in the linked sources; this entry does not document independent Aymorphic testing.
Key features
- Zero-output-token decision models: single forward pass returns a calibrated yes/no, choice, or score instead of generated text
- d1-3B: 3.1B parameters, multimodal (text + image), ~8ms latency on an RTX 4090
- d1-omni-600M: 600M parameters, multimodal (text + image or text + audio), tuned for low-resource/edge devices
- Scored 48.57 on Decision Index v0.2.1, ahead of the 12x-larger Decider-35B-A3B (47.11)
- Open weights on Hugging Face under the LFM Open License v1.0
- d1-ultra (Oct 15, 2026): closed hosted API, multimodal, $0.042 per million input tokens
Use cases
- Low-latency agent routing and guardrail checks (allow/deny, classify, score) without full generation
- On-device or edge classification tasks where generative-model latency is too high
- Structured decision steps inside larger agentic pipelines
Pricing
Free, open weights on Hugging Face; LFM Open License v1.0 permits free commercial use only for companies earning under $10M/year in revenue. Hosted d1-ultra API: $0.042 per million input tokens (input-only billing).
Pricing reflects the source record and may change. Confirm current terms with the provider.
Dates and sources
- Released
- 7 Oct 2026
- Record updated
- 18 Oct 2026
- Record last checked
- 18 Oct 2026