While tech leaders continue to debate the merits and pitfalls of open-weight AI models, developers aren’t slowing down. LTX on Tuesday released the newest version of its open world model, called LTX-2.5.

LTX-2.5 is an open world model, which is different from most AIs you probably know. While LLMs are good at generating text, world models are complex webs that translate the physical world into a language that generative AI can understand. The idea is that the model has a grounded understanding of our world, so it hallucinates less. Think about robots that can navigate a room without bumping into anything, or AI video clips that know to have people walk around objects rather than fade through them like a poltergeist. You can use world models to simulate digital environments, generate content and help program robots.

The new model has improvements to reduce the number of visual artifacts (or errors) the AI makes, better consistency across generations thanks to a new pixel diffusion process. Because it’s an open model, companies can download and customize the model for their own needs; that helps keep their intellectual property safe. So a movie studio could use LTX-2.5 to create and edit video clips featuring their characters. LTX-2.5 is optimized for local inference on Nvidia’s RTX GPU, which means developers can run it on their own hardware and keep their work away from cloud-based AI services.

LTX-2.5 is available to download and use now on sites like HuggingFace and ComfyUI.

LTX is one of the most popular developers of open-weight AI models. LTX’s models have been downloaded more than 33 million times by HuggingFace users, including those who made their own versions based on LTX’s blueprints. OpenAI and Google each have a family of open-weight models, GPT-OSS and Gemma 4, respectively, though the companies’ number of closed models far outnumbers their open-weights ones. Anthropic has no open Claude models.

Open-source advocates have long argued that the best way forward is for tech companies to make public disclosures about how their tools work. That debate has only intensified in the AI era. Popular Chinese models are being released in open formats, like the new Kimi 3 from Moonshot, which gives developers, researchers and cybersecurity experts insights into how it works. When the Trump administration was reportedly considering banning Chinese and other foreign AI models in July because it was worried about cybersecurity threats, many industry leaders spoke up, saying such a move could actually threaten cybersecurity.

While most AI leaders say open-weights models shouldn’t be banned, the big players aren’t rushing to release many of these kinds of models. Meta CEO Mark Zuckerberg is encouraging more open AI. He published a lengthy essay on Monday outlining his position that open-weight models will bring the “benefits of AI” to more people. Meta also released a new open-weights model called Muse Glimmer and promised to release the weights of another recent model, called Spark 1.2.

LTX has a similar philosophy. Neither government nor private companies would be good stewards of private AI models, CEO and co-founder Zeev Farbman wrote in a recent post. One is primarily motivated by ever-shifting political goals and the other by pure profit.

“It shouldn’t sit with a small number of decision makers, regardless of whether they’re in government or a limited number of powerful companies,” Farbman writes. “Decisions belong with the end users, through models they can control and inspect for themselves.”

Katelyn Chedraoui

Reporter 2

Katelyn is a reporter with CNET covering artificial intelligence, including chatbots, image and video generators. Her work explores how new AI technology is infiltrating our lives, shaping the content we consume on social media and affecting the people behind the screens. She graduated from the University of North Carolina at Chapel Hill with a degree in media and journalism. You can reach her at kchedraoui@cnet.com. See full bio