Mistral's Sovereignty Play

Only yesterday I was discussing Mistral off-work and pondering what they might be up to, given both their EU lobbying – which fits rather neatly with the European AI pitch I wrote about in April – and the fact that they haven’t been doing a lot of actual model releases (in practice, nothing really new, or at the same pace as other labs). I guess that they are positioning themselves squarely in the sovereignty arena, where a few other European companies have already started providing generic open-weight models for privacy- and US-sceptic customers that don’t want to be dependent on the whims of non-EU administrations. It’s a decent market to be in in Europe, but feels too regional, and I hope they get back to improving their own models, which are substantially lagging behind by any benchmark (if you care about benchmarks).