yolov5 vs cosmos

Two of the top projects in Open models, side by side: who leads on each score part and who each one suits, then setup, license, activity and what each review found.

#77th of 117 in Open models

yolov5

PyTorch YOLOv5 object detection, segmentation and classification models with training code

#1111th of 117 in Open models

cosmos

Open world foundation models for robotics and driving: video, reasoning, policies

Verdict

yolov5 ranks higher, 76 to 72. Of the 4 score parts we measure for both, yolov5 leads on 3, cosmos on 1.

  • Adoptionyolov5 leads by 24 points, 97 to 73: a wide gap.
  • Freshnesscosmos leads by 30 points, 100 to 70: a wide gap.
  • Maintenanceyolov5 leads by 6 points, 63 to 57.
  • Agent-readyyolov5 leads by 40 points, 70 to 30: a wide gap.

Pick yolov5 if you are

Engineers training or deploying YOLO detection models on custom image data

  • Five model sizes, from 1.9M to 86.7M params, with published CPU and V100 latency
  • detect.py accepts webcam, video, screen, YouTube and RTSP/RTMP/HTTP sources

Pick cosmos if you are

Robotics and autonomous-vehicle teams building on open world models

  • Three sizes: 64B Super, 16B Nano, 4B Edge for Jetson and RTX Pro 6000
  • Runs on Diffusers, Transformers, vLLM, vLLM-Omni, SGLang, TensorRT-LLM and NIM
yolov5 vs cosmos: score parts and facts
What we compareyolov5cosmos
Score parts, out of 100
Adoption97, widely used73, popular
Freshness70, recent100, active
Maintenance63, fair57, fair
Agent-ready70, partly30, minimal
Facts from GitHub and the README
Stars58.3k12.4k
LicenseAGPL-3.0 (copyleft)custom license (read the license)
Last commitNov 2026Nov 2026
Last releaseNov 2022Jun 2026
LanguagePythonJupyter Notebook
DockerNoNo
GPUOptionalRequired
arm64 or Apple SiliconNot statedNot stated

yolov5

YOLOv5 provides pretrained COCO checkpoints (n/s/m/l/x, plus 1280px P6 variants) and PyTorch scripts for detection, instance segmentation and classification. detect.py runs on webcams, files, folders, screen capture, YouTube and RTSP/RTMP/HTTP streams, train.py trains on custom data, and models can be exported to ONNX, TensorRT, CoreML and TFLite. The README points newer work to the separate ultralytics package.

Who it is for: Engineers training or deploying YOLO detection models on custom image data

Strengths

  • Five model sizes, from 1.9M to 86.7M params, with published CPU and V100 latency
  • detect.py accepts webcam, video, screen, YouTube and RTSP/RTMP/HTTP sources
  • Export to ONNX, TensorRT, CoreML and TFLite, with a Jetson deployment guide
  • Training extras: AutoBatch, multi-GPU, hyperparameter evolution, frozen layers, ClearML and Comet logging

Weaknesses

  • Last tagged release is v7.0 from 2022-11-22
  • AGPL-3.0 license; commercial use needs a paid Enterprise License
  • No pose estimation or oriented boxes (OBB); README directs those to the ultralytics package
  • COCO training takes about 1 to 8 days per model on one V100
  • GPU optional
  • Needs Python>=3.8, PyTorch>=1.8
  • Models: PyTorch .pt, ONNX, TensorRT, CoreML, TFLite

cosmos

Landing repo for Cosmos 3, a family of omnimodal world models with a Reasoner (text and vision to text) and a Generator (text, vision, audio, action to vision, audio, action). Ships three base sizes (Super 64B, Nano 16B, Edge 4B) plus post-trained checkpoints for driving simulation, transfer and DROID robot policies. The repo holds a cookbook of notebooks; training, curation and evaluation live in separate Cosmos repos.

Who it is for: Robotics and autonomous-vehicle teams building on open world models

Strengths

  • Three sizes: 64B Super, 16B Nano, 4B Edge for Jetson and RTX Pro 6000
  • Runs on Diffusers, Transformers, vLLM, vLLM-Omni, SGLang, TensorRT-LLM and NIM
  • Post-trained checkpoints for driving simulation, multi-view transfer and DROID robot policy
  • Companion repos cover SFT, LoRA, RL, distillation, data curation and evaluation

Weaknesses

  • Target hardware is H100, H200, B200, Jetson or RTX Pro 6000 class GPUs
  • Quickstart needs gated access to the Cosmos-1.0-Guardrail repo on Hugging Face
  • License is OpenMDW-1.1, not a common OSI license; custom licensing by email
  • README warns of artifacts in long, high-resolution or physically complex outputs
  • GPU required
  • Needs Hugging Face account and token, Cosmos-1.0-Guardrail (gated), PyTorch, diffusers
  • Models: Cosmos3-Super 64B, Cosmos3-Nano 16B, Cosmos3-Edge 4B, Diffusers, Transformers

More in Open models

Building with these?

We help teams pick an open-source AI stack and run it in production. Client work never changes a rank.