d1-3B
Released Oct 7, 2026 · knowledge cutoff unpublished
- Status
- Unknown
- Location
- United States
- Type
- System One
- Modality
- Multimodal
- Context window
- 33K
- Max output
- —
- Speed
- —
- Price ($/MTok in / out)
- — / —
- Cost per task
- —
- Open weights
- Yes
News2 stories
Strengths & weaknesses
Strengths
- Answers in one forward pass with no generated tokens, so a routing or moderation check costs a single prefill rather than a decode loop
- Judges images as well as text in the state, so screenshots and photos can be classified without a captioning step first
- Returns calibrated probabilities for yes/no, pick-from-options and ordered-score questions, so a confidence threshold can decide what escalates to a larger model
- Sized for edge hardware — Liquid quotes millisecond-scale latency on a single consumer GPU and ships an INT8 variant for newer NVIDIA devices
Weaknesses
- Cannot write text at all — Liquid's card is explicit that it is not a chat model, so anything needing an explanation has to go elsewhere
- Vision decisions are unbenchmarked — Liquid held back its private vision split and only checked that the backbone's vision carried over
- Free commercial use stops at US$10M annual revenue under the LFM Open License, and above that any production use needs a paid licence from Liquid
Not an LLM: an open-weights "System One" decision model in the same mould as Jev and Laya, 3.12B parameters on Liquid's LFM2.5-VL-3B with a SigLIP2 NaFlex vision encoder, that takes text or image state plus typed questions and returns calibrated probabilities in one forward pass, so none of this site's benchmarks apply and every cell is null by design. Released on Hugging Face with d1-omni-600M on 2026-10-07 under the LFM Open License v1.0, with a 32,768-token context per the model card; Liquid's seven-dataset text mean of 82.9 is its own measurement. Availability is self-host: Liquid's hosted decision API serves a separately announced model called d1 (2026-10-05, not on this site), and no hosted provider served this checkpoint at release. No price, knowledge cutoff or Artificial Analysis entry; sources are search excerpts of liquid.ai/blog/d1-open, huggingface.co/LiquidAI/d1-3B, docs.liquid.ai and liquid.ai/lfm-license, not pages read directly.
Benchmarks0 of 10 reported
MMLU-Proretired
GPQA Diamond
SWE-bench Verified
SWE-bench Pro
Terminal-Bench 2.1
AIME (latest)retired
Humanity's Last Exam
LMArena Elo
GDPval-AA v2
ARC-AGI-2
For developers
API model strings
Not researched
Licence
LFM Open License v1.0 · Restricted · commercial use permitted
Retirement
No retirement announced
Lineage
No recorded predecessor