Ultralytics v8.3.245 is live! 
Quick overview: Ultralytics v8.3.245 improves training/run telemetry by reporting system hardware capacity (RAM + disk) and logging best model file size at training end, plus a small docs footer refresh. No accuracy or training-logic changes 
New Features
System metrics added to platform callbacks
_get_environment_info() now includes:
- Total RAM (GB) (via
psutil) - Total disk capacity (GB) (via
shutil.disk_usage("/"))
This helps explain differences in caching behavior, performance, and out-of-memory issues across machines 
Improvements
Best model artifact size captured at training end
On on_train_end(), we now record the best checkpoint file size (modelSize) and attach it to the final training-complete payload.
Useful for:
- storage planning in CI / telemetry workflows
- comparing variants and checkpoints
- deployment constraints on edge devices
Docs / Maintenance
- Docs footer year updated 2025 → 2026 (MkDocs)
- Minor comment-only clarification for tensor shape annotations (no runtime behavior change)
Pull Requests Included
- Shape comment fix (no behavior change) by @raimbekovm in PR #23094
- Telemetry + docs footer updates by @glenn-jocher in PR #23095
You can review the full diff in the full changelog comparison, and details in the v8.3.245 release page.
Try it & share feedback
Update with:
pip install -U ultralytics
If you run into anything unexpected (or want additional telemetry fields), reply here with your OS, Python version, ultralytics version, and a minimal repro. Thanks for helping improve YOLO for everyone 