{"id":40503,"date":"2025-07-02T07:36:52","date_gmt":"2025-07-02T14:36:52","guid":{"rendered":"https:\/\/www.yardi.com\/blog\/?p=40503"},"modified":"2026-02-27T02:26:16","modified_gmt":"2026-02-27T10:26:16","slug":"open-source-ai-models","status":"publish","type":"post","link":"https:\/\/www.yardi.com\/blog\/open-source-ai-models\/","title":{"rendered":"4 companies pushing AI forward right now"},"content":{"rendered":"
<\/figure>\n

Welcome to the new and improved Real AI State Newsletter, now a weekly blog article! This week, we\u2019ll talk about the latest\u00a0open-source AI models<\/a>, including one\u00a0from\u00a0Mistral<\/a>, a French AI company. This innovative solution opens up an interesting conversation about combining smaller open-source models that can run locally (and hence inexpensively) with the large state-of-the-art models that tackle harder problems but at significant cost.<\/p>\n

Google, as usual, has been quite active and has two big announcements this week including\u00a0Gemini CLI<\/a>, their command line interface for writing computer code, and\u00a0Search Live<\/a>\u00a0with voice input in the Google app for Android and iOS. Anthropic won a critical court case where a\u00a0federal judge ruled<\/a>\u00a0that their use of books for training Claude qualifies as fair use and Tesla launched its\u00a0Robotaxi<\/a>\u00a0<\/a>service in Austin, Texas, showing the power of AI to tackle problems in the physical\u00a0world.<\/p>\n

With all this in mind, let\u2019s dig into these open-source AI models and see what they\u2019re all about.<\/p>\n

1. Mistral<\/h2>\n

What happened<\/h3>\n

Mistral released their\u00a0Small 3.2 open-source model<\/a>. It\u2019s an incredibly powerful, yet quite small LLM that can easily run on a standard laptop.\u00a0<\/p>\n

Why it matters<\/h3>\n

For those who don\u2019t\u00a0know, LLMs come in different sizes. The largest models that you may have heard of include OpenAI\u2019s GPT 4.5, or Google\u2019s Gemini 2.5 Pro which have well over a trillion parameters. That means that there are a trillion different elements that define the relationships between words and phrases (technically \u201ctokens\u201d) in the model. Generally, the more parameters a model has, the more \u201ccomputational horsepower\u201d it has to try and be intelligent. Mistral Small 3.2 only has 24 billion parameters, making it approximately one thousand times smaller than the big boys. Despite being so much smaller, it can outperform GPT 4 from May of 2024 on many benchmarks. This goes to show the incredible pace of development LLMs are experiencing.<\/p>\n

More importantly, Mistral Small 3.2 can be run locally, essentially for free, whereas GPT 4o is one of the more expensive models. For those of you who are trying to figure out a good AI strategy, one of the things to consider\u00a0is which types of AI are useful for which tasks and what the cost implications\u00a0are. Here at Yardi, we are working on building cloud-based tools for our clients using a variety\u00a0of AI models, taking advantage of their relative strengths as well as costs. Smaller, open-source AI models will definitely factor into our strategy. Use the cheaper, less capable models for functions that don\u2019t require extreme intelligence and then reach out to the expensive, state-of-the-art AI models when deep reasoning is needed.<\/p>\n

2. Google<\/h2>\n

What happened<\/h3>\n

Google continues to be a major player in the world of AI. This week there were two releases of note. First, they released\u00a0Gemini CLI<\/a>, another tool in their arsenal of AI powered software engineering. Google had already released\u00a0Jules<\/a>, their AI Coding Agent, but Gemini CLI now provides developers with direct access to the Gemini model from within their terminal, making it ideal for command-line based workflows.<\/p>\n

Second, they released\u00a0Search Live<\/a>, an amazing new interactive voice mode that is very similar to ChatGPT\u2019s Voice Mode but focused on search. Chatting with Google seems like a much better interface than typing queries into a text box.<\/p>\n

Why it matters<\/h3>\n

While most of you are probably not software engineers, you should be aware that there is a monumental shift occurring in software development right now. With the rise of tools like Google\u2019s\u00a0Jules\u00a0and\u00a0Gemini CLI, as well as\u00a0Codex\u00a0<\/a>from OpenAI,\u00a0Claude Code\u00a0from Anthropic,\u00a0Cursor, and\u00a0Github Copilot, we\u2019re seeing the rise of the AI Software Engineer. In fact, it is likely that within the next few years, most lines of computer code will be written by AI rather than a human. This means that the cost of developing software will plummet and the number of domain experts who can build\u00a0software will explode. If you really want to learn more about what \u201cSoftware 3.0\u201d looks like, watch\u00a0this talk by Andrej Karpathy. The business implications are staggering.<\/p>\n

\n
\n