Palo Alto AI初创公司Elorian推出首个AI模型Glimpse，引入了不同于传统方法的图像处理机制1。与常见的将图像转换为tokens后进行推理的做法不同，Glimpse能在推理过程中重复查看图像、裁剪特定区域和抽取视频帧1。该公司声称该模型在科学图表推理和短视频字幕生成等任务上超越了GPT-6 Astra1。
Elorian已于种子轮融资中筹得5500万美元，估值达3亿美元1。投资方包括Striker Venture Partners、Menlo Ventures、Altimeter Capital，以及Nvidia和Google首席科学家Jeff Dean参与其中1。公司CEO Andrew Dai曾任职Google DeepMind研究员1。Glimpse的定价为输入端每百万tokens 60美分、输出端每百万tokens 3美元1。该模型采用数百个GPU进行训练1。
Palo Alto-based AI startup Elorian has launched its inaugural model, Glimpse, featuring a novel approach to visual reasoning that sets it apart from conventional image-to-token processing methods 1. Rather than converting images into tokens before conducting analysis, Glimpse repeatedly examines images, crops specific regions, and extracts video frames throughout its reasoning process 1.
The company has priced its API access at $0.60 per million input tokens and $3 per million output tokens 1. Elorian secured $55 million in seed funding at a $300 million valuation, with backing from investors including Striker Venture Partners, Menlo Ventures, and Altimeter Capital, alongside participation from Nvidia and Google Chief Scientist Jeff Dean 1. The model was trained on several hundred GPUs 1.
CEO Andrew Dai, a former Google DeepMind researcher, leads the startup 1. According to the company's claims, Glimpse outperforms GPT-6 Astra on tasks including scientific chart reasoning and short-form video captioning 1.
评论
还没有评论，欢迎留下第一条。