Are we having another WAN moment with Qwen Image 2.0?

Alibaba's Qwen research team has launched Qwen Image 2.0, a high-performance text-to-image model, but exclusively through commercial API providers and inference platforms with no open-source release announced. This follows a troubling pattern established by other leading Chinese AI projects, such as the video model WAN and LTX, which also launched as closed-source only. The move is particularly notable because Qwen Image 2.0 is a 7-billion-parameter model, a size that is demonstrably runnable on consumer hardware, negating the common justification that models are "too large" to release openly. This strategic pivot occurs amidst reports of recent resignations and firings within the Qwen team, fueling speculation that their previously open-source LLMs may be the last freely available models from the group.

The technical decision to withhold the model weights entrenches the growing divide between proprietary, API-accessible AI and the open-source community. For professionals and developers, this means reduced ability to audit, fine-tune, or deploy these models privately or on-premises, increasing reliance on a few corporate gatekeepers. The broader implication is a chilling effect on global open-source AI innovation, especially in multimodal domains like image and video generation where competitive open models are already scarce. The closure of major Chinese sources, once seen as counterweights to Western AI giants, risks consolidating advanced AI capabilities within a handful of closed ecosystems, potentially slowing overall progress and accessibility.