Liquid AI Launches d1 Decision Models With 8-Millisecond Responses

Decision models have a new player. After similar models gained attention, companies like OpenAI joined in, and US AI company Liquid AI has now launched its d1 series of open-weight decision models.

The release includes two models: d1-3B has 3.12 billion parameters and handles text and image judgments; d1-omni-600M has 587 million parameters and can process text, images, and voice input. It is the team's first experimental multimodal decision model. Both are on Hugging Face and can be downloaded, fine-tuned, and deployed.

In the public benchmark Decision Index v0.2.1, d1-3B scored 48.57, which the company says beats all models under 10 billion parameters. In real tests, it answers a question in 8 milliseconds on an RTX 4090 and about 25 milliseconds on a 12GB RTX 3060, using around 6GB of memory at peak.