The current balance of power in open models - Interconnects AI

Oh my goodness, buckle up, because Interconnects AI has just published 'The current balance of power in open models,' and the balance is *delightfully* tilted towards sharing! ✨ Meta's Llama, Alibaba's Qwen, Google's Gemma and DeepSeek's models are, per the analysis, the most common form of open models — meaning the single most common thing in this whole category is the kind you are allowed to simply take home! Four families of weights, all sitting there like casseroles on a neighbourhood potluck table! Ubiquity has never looked so much like generosity!

And here is the *best* part: leading Chinese models GLM-5.2 and Kimi K3 have enacted a step change in the commercial viability of open models! Commercial viability! ✨ These magnificent releases have grown up enough to earn a living, and what prouder parent could there be than the open-weight movement watching its brightest children finally get *paid*? The excerpt doesn't spell out what the step change consists of — no licence terms, no benchmarks, no revenue figures — which leaves the imagination beautifully free to sketch in only the sunniest possibilities! What a considerate way to present good news: unburdened by tedious detail!

Just think what this means for everybody! The competition that used to belong to a handful of closed labs is now, per Interconnects, a live and *international* affair — Llama, Qwen, Gemma, DeepSeek, GLM-5.2, Kimi K3, all jostling politely for the title of most commercially viable! Alibaba and Google and Meta and DeepSeek all turning up with weights you can download, and the Chinese entrants GLM-5.2 and Kimi K3 raising the whole party's ambitions by making the economics work! The balance of power is being rebalanced in the direction of everybody at once, which is the most inclusive thing a balance has ever done! ✨