Editorial illustration for Liquid AI's LFM2.5-VL-450M: model with bounding boxes, sub‑250 ms inference
Liquid AI's 450M Model: Ultra-Fast Vision-Language Tech
Liquid AI's LFM2.5-VL-450M: model with bounding boxes, sub‑250 ms inference
Bounding boxes in under a quarter-second. That’s the headline. Liquid AI’s LFM2.5-VL-450M doesn’t just see, it pinpoints, drawing boxes around objects in images with inference times that slip below 250 milliseconds.
At 450 million parameters, this model is built for edge deployment, yet it doesn’t force a one-size-fits-all speed. Users can dial the number of image tokens and tile count on the fly, trading latency for quality without retraining. That flexibility meets hardware where it lives, from a phone to a server.
The training story is just as deliberate: the team scaled pre-training from 10 trillion to 28 trillion tokens over the previous iteration, then sharpened the model with preference optimization and reinforcement learning. The result is a vision-language system that follows instructions more reliably, grounds objects with precision, and now, for the first time, predicts bounding boxes. Multilingual support rounds out the package.
This isn’t a marginal update, it’s a redefinition of what a compact model can do.
Liquid AI just released LFM2.5-VL-450M, an updated version of its earlier LFM2-VL-450M vision-language model.
A model that draws boxes under 250 milliseconds isn’t just fast, it redefines what’s possible on the edge. Liquid AI has given developers a dial, not a fixed lens, letting them choose between pixel-rich understanding and raw speed without starting over. The bounding box capability is the headline, but the real story is the training discipline: 28 trillion tokens, then careful post-training to glue language to vision with reliability.
That’s the difference between a demo and a deployable system. For anyone building in resource-constrained environments, this is the kind of precision that doesn’t waste a millisecond.
Related Reading
LLMs & Generative AI ChatGPT's 'Nerdy' tweak rewards goblin metaphors in answers, study finds
LLMs & Generative AI Google tests visual 'magazine-style' UI for Gemini 3 Pro users
LLMs & Generative AI AI Engineers Face Rising Costs, Need New Strategies for Efficiency
From Archives Sarah says plugging memory into a harness is like plugging driving into a car
From Archives AI models spam help requests if rewards equal correct answers, accuracy 5.4%
Common Questions Answered
How does the LFM2.5-VL-450M model achieve sub-250 ms inference speed?
The model allows developers to dynamically adjust image processing parameters at inference time, such as maximum image tokens and tile count. This flexibility enables a speed/quality tradeoff without requiring retraining, making it adaptable to different hardware compute budgets.
What recommended generation parameters does Liquid AI suggest for the LFM2.5-VL-450M model?
For text generation, Liquid AI recommends using a temperature of 0.1, min_p of 0.15, and a repetition penalty of 1.05. For vision inputs, they suggest setting min_image_tokens to 32, max_image_tokens to 256, and enabling image splitting.
What are the key capabilities of the LFM2.5-VL-450M vision-language model?
The model can predict bounding boxes, follow instructions more closely, understand multiple languages, and invoke functions. It is designed to fit on edge devices like NVIDIA's Jetson Orin, AMD's Ryzen AI Max+ 395, and Qualcomm's Snapdragon 8 Elite, with a target inference speed of under 250 milliseconds.
Further Reading
- LFM2.5-VL-450M: Structured Visual Intelligence, Edge to Cloud — Liquid AI Blog
- LiquidAI/LFM2.5-VL-450M — Hugging Face
- LFM2.5-VL-450M — Liquid AI Docs
- LFM2 Technical Report — arXiv
