> ## Content Index > Fetch the complete content index at: https://itsfoss.com/llms.txt > Use this file to discover other available public pages before exploring further. # A New Open-Source AI Model 'Molmo' Joins The Race to Tackle The Big Tech - URL: https://itsfoss.com/news/molmo-open-source-ai-model/ - Published: 2024-09-26T10:54:19.000Z - Updated: 2024-09-26T10:54:19.000Z - Description: The list of open-source AI models is growing, and that's a good sign! - Author: Sourav Rudra - Tags: AI Spotlight, News, #Import 2025-11-08 21:15 Recently, the field of AI has been making some significant strides when it comes to following an open-source approach to development. The proprietary crowd wished they had that kind of close-knit community of developers working towards achieving great things. In the space, [multimodal models](https://en.wikipedia.org/wiki/Multimodal%5Flearning?ref=itsfoss.com) have been a major point of interest due to their powerful capabilities, which enable them to process many types of input data, such as text, images, and videos, resulting in richer, more context-aware outputs. There is always someone or some organization trying to offer a model that is more powerful and tractable than the ones that came before. In this instance, we have a Seattle-based non-profit AI research institute called [Ai2](https://allenai.org/?ref=itsfoss.com) introducing a new family of open-source multimodal models called **Molmo**. Let's check it out. πŸ˜ƒ ## Molmo: Better Than Proprietary Models? Introduced as **a family of state-of-the-art (*SOTA*) multimodal AI models**, Molmo can be used to develop things like a chatbot, interactive educational platforms, content moderation systems, or even AI agents. As demonstrated by the video above, you could ask a Molmo-powered agent to tell you a joke, help you navigate a new area, convert ideas from a whiteboard into a digital document, and carry out many such everyday tasks. Molmo has been trained on a dataset called **PixMo**, which **consists of almost 1 million carefully chosen image text pairs** with two main categories of data. One is the dense captioning data used in multimodal pre-training, and the other is supervised fine-tuning data, used for enabling a wide range of user interactions. Molmo consists of **four different models** with varying capabilities: - **MolmoE-1B** - **Molmo-7B-O** - **Molmo-7B-D** - **Molmo-72B** In this lineup, the 72B is the flagship model, the 7B-O is the most open one, and the 7B-D is the demo model, with the E-1B being the most efficient one. Here, the B's represent the billion parameter metric. πŸ“‹ The Molmo 72B and 7B-D's LLM backbone consists of some closed-source data and code. To demonstrate the performance of Molmo, **the developers showcased two benchmarks** where the four Molmo models were pitched against other, more popular/proprietary models. ![an illustration showing how molmo performed against other models in eleven academic benchmarks and human preference elo rating](https://itsfoss.com/content/images/2024/09/Molmo_bench.png) Molmo evaluated against other options. On both benchmarks, the Molmo family of models was able to either perform better, or be close in performance to the likes of GPT-4O, Gemini 1.6 Pro, Claude 3.5 Sonnet, Qwen VL2 72B, and Pixtral 12B. To see how it works, **I tested out the demo Ai2 has provided**. It was running the Molmo-7B-D model, which was only available for [vision](https://en.wikipedia.org/wiki/Computer%5Fvision?ref=itsfoss.com)\-related tasks, and required me to upload an image before it would accept a prompt. ![a screenshot of molmo explaining what type of cat is in a photo](https://itsfoss.com/content/images/2024/09/Molmo_test_a.png) Pic Source: [Pexels](https://www.pexels.com/photo/low-angle-shot-of-a-tabby-cat-208984/?ref=itsfoss.com) First, I asked it to identify what kind of cat the photo I uploaded had on it. It was able to correctly point out that it was a domestic shorthair cat with tabby markings. It seemed to pull in information from [Wikipedia](https://en.wikipedia.org/wiki/Tabby%5Fcat?ref=itsfoss.com) and some other sources. ![a screenshot of molmo explaining what is happening in a photo](https://itsfoss.com/content/images/2024/09/Molmo_test_b.png) Pic Source: [Pexels](https://www.pexels.com/photo/photo-of-people-gathering-in-room-2833037/?ref=itsfoss.com) Then, I asked Molmo what was happening in a photo filled with people. It gave me a very relevant answer that it was a large-scale event (*unsure about the church service part*) with a stage and many people gathered; it even described what the woman in the focus of the photo was wearing and what she was doing. But, these were some basic questions I asked Molmo to answer. You can ask it for more advanced things like explaining a graphic, fact-checking a photo to ascertain whether it is fake or not, and more such requests. ### Want To Try It Out? There is no requirement to get an API or sign up for a subscription to try Molmo. The developers intend to provide all of their model weights, captioning/fine-tuning data, and source code in the near future for everyone to use. But, for now, you can check out [the demo](https://molmo.allenai.org/?ref=itsfoss.com) for Molmo (*vision-related tasks only*). There is also a [Hugging Face](https://huggingface.co/allenai?ref=itsfoss.com) page for those who want to get started with any one of those models on their hardware. [Molmo (demo)](https://molmo.allenai.org/?ref=itsfoss.com) For more details on Molmo, you can go through the official [blog](https://molmo.allenai.org/blog?ref=itsfoss.com) and [paper](https://molmo.allenai.org/paper.pdf?ref=itsfoss.com) to get an overall view of things. *πŸ’¬ In these AI-crazed times, which open-source AI model has caught your attention?* **Suggested Read** πŸ“– [Finally! Brave Brings Local AI to the BrowserA step in the right direction to protect users’ privacy.![](https://news.itsfoss.com/content/images/size/w256h256/2022/08/android-chrome-192x192.png)It's FOSS NewsAnkush Das![](https://news.itsfoss.com/content/images/2024/09/brave-browser-brings-local-ai.png)](https://itsfoss.com/brave-browser-local-ai/)