Two Western companies are releasing open-weight AI models tailored toward businesses and governments that they claim are as powerful as the leading models offered by China—and could eventually rival some of the biggest closed models from US-based tech giants. Open weight? Is this about boxing? No, actually. Open weight means that, unlike with closed weight models, users can download and customize them with their own proprietary data. They are also cheaper to run than the difficult-to-budget closed models, which could present a new challenge for US-based AI leaders Anthropic, Google, and OpenAI. The new playersUS-based Reflection AI, which Nvidia financially backs, is launching its first model, Beam, which it says is adept at coding and agentic tasks. Meanwhile, France-based Mistral is debuting Mistral Large 4, called “Le Chonk” due to its 1 trillion parameters...which is a chonky (industry term) amount of parameters. Should these new models have the capabilities claimed by their companies, they could be massive disruptors to the status quo: - Businesses and governments in the West are currently hesitant to use the Chinese open models due to security concerns. Reflection and Mistral could be the answer.
- OpenAI and Anthropic—already facing growing security concerns—may have new competition as they prepare to go public.
Speaking of security: By putting powerful models that anyone can tweak into circulation, it creates the potential for even more cybersecurity issues. But Reflection and Mistral see another side of the coin: Mistral VP of Science Pierre Stock told Axios that more availability of models can help improve defenses and allows researchers to audit the technology. More to come: Reflection will release full technical details of Beam later this month. Mistral’s Le Chonk, which is currently only being previewed for developers, will have a wider release on Oct. 27 after more testing, per Axios.—DL | 
