v0 adds open-weight models, starting with Kimi K3
By Steven Van ·
Moonshot AI's 2.8-trillion-parameter model is now selectable in v0.app alongside v0's closed frontier options.
v0 now supports open-weight models, a new category alongside the closed frontier models it's offered so far. The first model in is Kimi K3 from Moonshot AI, available now at v0.app/?kimi3.
Why Kimi K3 is a notable first pick
Kimi K3 isn't a token open-source gesture, it's currently the largest open-weight model publicly available: a 2.8-trillion-parameter mixture-of-experts model from Beijing-based Moonshot AI, natively multimodal with a 1-million-token context window. In blind developer testing on Arena's Frontend Code evaluation, K3 ranked first, ahead of Claude Fable 5, and Moonshot reports frontier-level performance across coding, agentic, reasoning, and knowledge tasks more broadly, trailing Fable 5 and GPT-5.6 Sol only slightly overall. For a UI-generation tool like v0, strong frontend coding performance is exactly the benchmark that matters most.
What 'open-weight' changes for v0 users
Open-weight models can be inspected, fine-tuned, and in principle self-hosted, a different cost and trust profile than a closed API-only model. For v0, adding an open-weight option means users get a genuinely competitive alternative to closed frontier models without the licensing and pricing structure that comes with them, generally cheaper to run given the model can be served more flexibly than a proprietary one.
Part of a broader open-weight moment
Kimi K3's release fits a wider trend of Chinese labs pushing frontier-capable open-weight models as U.S. compute restrictions push a different strategy: give away the weights, compete on adoption instead of API margin. v0 picking Kimi K3 as its first open-weight model is a signal that these releases are now good enough to sit next to closed frontier models in a mainstream coding tool's model picker, not just a research curiosity.
Sources: v0 on X, Tom's Hardware.