Xiaomi open-sources Xiaomi-Lens-0, a 4B multimodal model that outputs pixel-level mask tokens mid-answer
Xiaomi's SeerRay-Lab has open-sourced Xiaomi-Lens-0, a 4B multimodal model built on Qwen3.5-4B. The model emits mask tokens for pixel-level image regions in the middle of an answer. The lab also released a benchmark called LensBench.
Xiaomi's SeerRay-Lab has open-sourced Xiaomi-Lens-0, a 4B-parameter multimodal model that emits mask tokens for pixel-level image regions in the middle of an answer, according to Pandaily.
What was released
- Model: Xiaomi-Lens-0, a 4B multimodal model.
- Base: Qwen3.5-4B.
- Behavior: The model produces mask tokens that identify pixel-level regions while it is generating a response, so the answer can point to specific areas of an image rather than only describing them.
- Benchmark: SeerRay-Lab also released LensBench alongside the model.
Details
The release comes from SeerRay-Lab, part of Xiaomi. Pandaily's headline describes the model as one that points to image regions pixel by pixel. Context window, pricing, benchmark scores and training cutoff date are not stated in the summary available at publication, so none are listed here.
What this means
Xiaomi-Lens-0 is a 4B model that produces pixel-level region output as part of its answer, and it is open-sourced together with LensBench, a benchmark released by the same lab. Developers working with image-grounded answers now have both a model and an accompanying benchmark from SeerRay-Lab to examine.
Related Articles
Xiaomi retires MiMo-V2.5-Pro and MiMo-V2.5 API models, moves service to MiMo-V2.6 family
Xiaomi discontinued MiMo-V2.5-Pro and MiMo-V2.5 at 10 a.m. Beijing time on October 21. Their API services now run on the MiMo-V2.6 family, and requests to the old model names have been routed to MiMo-V2.6-Pro and MiMo-V2.6-Flash since October 14.
iFLYTEK releases Spark X2.6, a 293B-A30B MoE model trained on domestic compute
At its 1024 developer festival, iFLYTEK released Spark X2.6, a 293B-A30B mixture-of-experts model trained on domestic compute. The company also open-sourced the on-device Spark X2.6-4B and debuted the SparkAudio 1.0 speech model.
NVIDIA releases Kumo Forecast, an Apache 2.0 time-series model built on a ~340M-parameter MOMENT-1-large encoder
NVIDIA has released Kumo Forecast (formerly NV-Tesseract Forecasting), a multivariate time-series forecasting model under Apache 2.0. It pairs a frozen MOMENT-1-large encoder of roughly 340 million parameters with a trainable forecasting head, optional nearest-neighbor retrieval over historical data (DARR), and a built-in interpretability framework.
NVIDIA releases Nemotron 3.5 Embed 8B, scoring 61.17 nDCG@10 on ViDoRe v3.1 among dense retrievers
NVIDIA has released Nemotron 3.5 Embed (8B BF16, 1B BF16, 1B NVFP4), Nemotron 3.5 Rerank 8B BF16 and Nemotron 3.5 Parse (631M parameters) as open, commercially available models. NVIDIA reports that the 8B embedder ranks first among dense retrievers on ViDoRe v3.1 with 61.17 nDCG@10, and that adding the reranker cuts agent token cost per query by roughly 13-14%.
Comments
Loading...