Show HN: Time Wizard (huggingface.co)

A new AI model called Time Wizard has been unveiled, achieving impressive accuracy in reading the time from photographs of analog clocks. Built on a 450 million parameter vision language model (LFM2.5-VL-450M) by Liquid AI, Time Wizard can interpret clock faces and output the time in JSON format. In tests using 200 held-out photographs, it performed similarly to the advanced GPT-5.6 Sol model, while significantly outperforming Claude Fable 5.1 and Claude Opus 5. Time Wizard’s ability to read the hour and minute accurately was demonstrated with a 59.0% success rate within one minute and 31.0% exact accuracy, all achieved with no elaborate reasoning, unlike its competitors which consumed a larger number of tokens in attempts to decipher the hands of the clock. This development is noteworthy for the AI/ML community due to its efficient design and lower resource requirements, with training costs estimated at just $0.15. Time Wizard leverages a streamlined model architecture and a specialized training dataset, allowing it to interpret clock angles through a single forward pass, rather than extensive reasoning processes. This efficiency indicates potential for applying similar methodologies in other vision language tasks, making Time Wizard a valuable addition to the growing collection of AI models specializing in visual perception and interpretation. The model is available under the LFM Open License v1.0, promoting accessibility for further research and development in this domain.