Elorian launches Glimpse, a model that ‘thinks in images’ and claims to match the performance of GPT-6 Astra at 6 per cent of the price

The Palo Alto-based start-up has unveiled a visual reasoning model that revisits its image representations whilst processing, charged at $0.60 per million tokens input.

Elorian, a start-up dedicated to multimodal intelligence applied to the physical world, unveiled Glimpse on October 16, which it describes as a "visual thinking model". Instead of converting an image into tokens once and for all, it reworks its internal visual representations as it is reasoned. The objective: to track moving objects, understand their relative positions, anticipate the physical consequences of an action or precisely count elements.

According to the company, this much smaller model equals or exceeds GPT-6 Astra and Claude Opus 5.5 on several multimodal benchmarks, with for example 92.90 in reasoning on scientific graphs, 83.36 in open-vocabulary counting and 42.70 on the detection of robotic manipulation events. Glimpse is priced at $0.60 per million tokens for input and $3 for output, which is 94% less than Astra, and can be accessed via API or via a chat interface with free trial access.