Roboflow
@roboflow
Build & use computer vision models fast ✨
Get started: roboflow.com
Open source datasets & models universe.roboflow.com
Joined June 2019
full evals are now live for grok 4.6 for sure a model to consider depending on your use case, cost, and latency you can compare results across models on playground so worth testing with your own data- had the chance to start evaluating grok 4.6 earlier this week, our vision eval is tough real world data only 4.6 is getting better at vision, a few examples comparing to 4.5 below quick takeaways - predicted boxes are closer to ground truth and tighter - predicts roughly the
- tracking is core to any video analytics use cases. you need the added accuracy for understanding complex scenes our trackers repo is bringing research papers into open source for everyone to use test out different trackers with this colab github.com/mirrash7/Exper… thanks toMcByte shipped in trackers 2.6.0 similar to ByteTrack, but association is guided by segmentation masks (SAM + Cutie), not just boxes when players overlap and box IoU gets ambiguous, masks settle who is who here it is side by side with OC-SORT link: github.com/roboflow/track…
- our deep dive on qwen3.8-max lots of examples and breakdowns by each category we eval, it's a top choice for vision tasks. we will eval open weights when they arrive we also added muse spark 1.2 and meta is a top vision model provider now. three way race between google, meta,Qwen3.8-Max blogpost - object detection (taxt and box prompts) - object counting - image reasoning - data extraction - cost and speed link: blog.roboflow.com/qwen3-8-max/ tl;dr below Qwen3.8-Max tops our object detection benchmark. it performs well across satellite images,
