Reflection AI and Mistral launch open models Beam, Le Chonk

The week's launches mark a shift in the Western open-source race, which both Reflection AI and Mistral framed as a bid to break China's open-source monopoly. The two labs set their sights not only on Beijing but on each other, with both claiming they beat other open Western models such as those from Thinking Machines and Nvidia. That emphasis on the intra-Western race is notable: US and European companies have shifted toward using more Chinese models over the past year, so a Western win means little without buyers.

Reflection CEO Misha Laskin offered one answer to that problem. Rather than trying to dislodge Chinese models from their foothold among developers and large companies, he told Semafor he sees a market among entities that can't or won't use Chinese models — heavily regulated industries and governments — which might otherwise have no choice but to pay for closed American AI. The timing supports the pitch. Businesses are entering year-end budget reviews with large AI spend figures to account for, governments are bracing for sticker shock as $1-a-year pilot deals expire, and new data from Ramp suggests companies are switching AI providers more than ever.

The open question is whether that market will accept less advanced models in exchange for affordability and control. The launches so far beat other open Western models but trail the top Chinese models and the leading closed systems from OpenAI and Anthropic on prominent benchmarks, so the appeal may hinge on buyers deciding that cost and control outweigh benchmark performance for some tasks.