SenseTime releases SenseNova U1.5 with native 4K generation

Background SenseNova U1.5 continues the SenseNova U1 line, which SenseTime had already released as an open-weight unified model. The U1.5 release extended that line with a focus on high-resolution image generation.

What happened SenseTime released SenseNova-U1.5-8B-MoT on 20 August 2026, after a preview on 31 July 2026 ^[1]. The preview had already shown the direction of the release, which was native 4K generation. The full release made the 8-billion-parameter model available as open weights.

Key details The 8B mixture-of-transformers model unifies understanding, reasoning and generation in one model ^[2]. It supports native image generation up to 4096x4096 pixels, which is the 4K resolution the preview focused on ^[2]. Native generation means the model outputs the image at its final size, not as a smaller image that is enlarged later.

SenseTime reported GenEval 0.92 and MMLU-Pro 86.67 for the model ^[2]. GenEval measures how well an image generator follows the details of a text prompt, and MMLU-Pro tests knowledge and reasoning across many academic subjects. These are the company's own reported results.

The technical report followed on 11 September 2026 ^[1]. The weights are released under the Apache License 2.0 ^[1].

Impact For readers and developers, the most concrete effect of the release is that a model of this size can be downloaded and run for high-resolution image generation, under a permissive license.

For readers new to this area, the basic idea is simple. A text-to-image model takes a written description and produces a picture, and a model that also reads images can check a picture against a description. When one network handles both jobs, the same learned knowledge can serve understanding and generation. Whether a release of this kind matters to a given reader depends on whether they need local control over the model, which open weights allow, or whether they only want to use it through a hosted service.