Claude models dominate a new benchmark for resisting Russian propaganda, while Mistral's models rank in the bottom third and struggle to detect disinformation.

What happened: The Institute of the Estonian Language released a benchmark testing 60 AI models on their susceptibility to Russian propaganda using 75 questions across three languages and 14 propaganda narratives. Anthropic's Claude Fable 5 scored 95.2 and leads the ranking, followed by Claude Opus 4.7, Nvidia's Nemotron 3, and Alibaba's Qwen 3.6 Plus. Mistral's models, including Medium 3.5, landed in the bottom third.

Why it matters: Russian networks deliberately feed AI systems millions of disinformation articles, and OpenAI recently shut down a Russian campaign using ChatGPT to spread propaganda before Germany's federal election. The benchmark reveals that some widely used models fail to spot and reject propaganda narratives. Mistral's poor results are particularly significant for a company that positions itself as a European alternative to US and Chinese providers.

What to watch: Mistral's misinformation rate of 36.67 percent, per a Newsguard study, comes as the company is currently negotiating a 3 billion euro funding round at a 20 billion euro valuation, while its flagship models already struggle to keep up with competitors.