[NEWS] AI2、Molmo, a new AI that surpasses ChatGPT, announced

Allen Institute for Artificial Intelligence (AI2) Unveils "Molmo," a Multimodal AI Model Surpassing GPT-4o in Benchmark Tests.

The Allen Institute for Artificial Intelligence (AI2) has announced "Molmo," a multimodal AI model that outperforms GPT-4o in benchmark tests. Molmo can process both text and images, and it excels in image recognition tasks.

It surpasses top-tier models like OpenAI's GPT-4o, Anthropic's Claude 3.5 Sonnet, and Google's Gemini 1.5 in benchmarks. Molmo, being multimodal, can process, understand, and analyze different types of data (modes) simultaneously. Like other major foundation models, it can accept and analyze images and files.

In a post on X (formerly Twitter), AI2 stated that Molmo uses "1,000 times less" data than its competitors. The model is a result of new challenges and technological advancements led by the company founded by Paul Allen and headed by CEO Ali Farhadi.

AI2 also posted videos on YouTube and social media demonstrating how Molmo can be used via smartphone. Users simply take a photo and send it to the AI, which rapidly analyzes the objects in front of them.

The AI can instantly count the number of people in a scene, determine whether menu items are vegan, analyze posters on lampposts to identify which bands play electronic music, or read and convert whiteboard content into graphs.

AI2 highlighted that this release underscores its commitment to open research by offering high-performance models with open weights and data to a broader community, as well as to companies seeking fully owned, controlled, and customizable solutions.

Molmo consists of four primary models with varying parameter sizes and functionalities.

These models have demonstrated superior performance across several third-party benchmarks, surpassing many proprietary alternative models. All of these models are available under a flexible Apache 2.0 license, making them suitable for both research and commercial applications.

One of the main models, Molmo-72B, received the highest academic rating, scoring the highest on 11 major benchmarks, and has garnered support from users, ranking just behind GPT-4o.

Reeference:Molmo
Image: Shutterstock

Related articles

Former staffer expresses concern over lack of proper safeguards and oversight in AGI development

Vice President Harris says she will promote crypto assets and AI