OPEN LINK ↗
// 1h agoMODEL RELEASE
Liquid AI d1-3B lands for edge decisions
Liquid AI released d1-3B, an open-weight 3B decision model that returns typed answers from text and images in one forward pass. It reaches 8 ms on an RTX 4090 and 50 ms on Jetson Orin Nano. ([official announcement](https://www.liquid.ai/blog/d1-open), [model card](https://huggingface.co/LiquidAI/d1-3B))
// ANALYSIS
d1-3B’s real innovation is replacing token generation with fast, calibrated decisions—a compelling fit for systems where latency matters more than conversation quality.
- –Scores 48.57 on Decision Index 0.2.1, leading models under 10B
- –Targets reranking, agent guardrails, visual inspection, and real-time classification
- –Multimodal text-and-image inputs preserve much of LFM2.5-VL-3B’s vision capability
- –Hugging Face, vLLM, SGLang, and quantized deployments lower adoption friction
- –The LFM Open License includes commercial-use conditions above its revenue threshold ([license](https://www.liquid.ai/lfm-license))
// TAGS
d1-3bsmall-llmopen-weightsinferenceedge-aimultimodalbenchmark
DISCOVERED
1h ago
2026-10-08
PUBLISHED
2h ago
2026-10-08
RELEVANCE
9/ 10
AUTHOR
LaronskiGeek