Qwen3.5-Max-Preview Goes Live on Arena
Alibaba's Qwen team has put Qwen3.5-Max-Preview on the Arena evaluation platform for public testing and plans to release the finished model within the next two weeks.
The Qwen team announced that Qwen3.5-Max-Preview is now live on Arena, the crowdsourced model-comparison platform that started as LMSys Chatbot Arena. The team says the model "has demonstrated exceptional performance during the preliminary evaluations" but gives no scores, rankings or details about those evaluations.
The model is still a preview. According to the post, the team is making "final optimizations" ahead of a release planned for the next two weeks. Until then, it is asking the community to test the model at arena.ai.
The announcement is brief. It gives no information about the model's architecture, parameter count, context length, modalities, pricing or license. It also doesn't say whether the full release will come with open weights or be available only through an API. In the past, the "Max" tier has been Qwen's flagship model and has not been released with open weights, but the post doesn't say whether that will change.
Putting a preview on Arena before launch lets developers collect blind, head-to-head human preference votes from real users. The team gave no Arena leaderboard placement. Arena usually waits until a model has enough votes before ranking it publicly.
The next things to watch are an Arena ranking and the full release with technical documentation, which the team says is about two weeks away.