Open AI models that companies can run themselves now beat or match paid rivals on some finance tasks, which could mean lower costs and more control over private data.
Open AI models, the kind a company can download and run itself, are catching up with the big paid ones. This week Moonshot AI released Kimi K3.1 Pro, and it beat OpenAI's GPT-6 Astra on a financial research test. Mistral released its Large 4 model the same way.
Releasing the weights means a company publishes the inner settings that make a model work. Others can then run it on their own systems. Moonshot and Mistral both pitch finance as a core use case. Tencent's Hy4 also won a blind test judged by 213 financial professionals.
Nvidia added a tool too. Nemotron 3.5 Transcribe turns call recordings into text in 40 languages on a single chip, so banks can keep customer audio in-house. Google also rolled out Genie 4, which builds simulated 3D worlds where robots and software agents can train.
The material does not say what these models cost to run, what the license terms are, or how Mistral Large 4 scored on any test. It also does not name the financial research test that Kimi won.
Test scores are only one guide. Artificial Analysis rebuilt its Intelligence Index so labs cannot train on most of the test. For anyone using AI on money tasks, the practical point is that cheaper options may now be worth a trial on your own work.
If you pay for AI on finance tasks, test an open model on your own work before assuming the paid one is better.
Mastercard describes AI tools that could buy things for you, and says trust is the hard part: who gets to spend your money, and under what rules.
Read more: Mastercard outlines how AI assistants could shop and pay for youWorld models, AI that learns physics from video, are being built into home robots and driverless cars, and your own videos may help train them.
Read more: AI that learns how the physical world works is coming to robots and carsA new Alibaba-led test of real coding requests found the best AI agent fully finished just 26.4 percent of tasks, so human review still matters.
Read more: AI coding helpers fully finish only about 1 in 4 real tasks
