OmniParser V2
Pure-vision screen parsing for GUI agents.
About
OmniParser V2 is Microsoft's February 2025 screen-parsing model for vision-based GUI agents: it detects interactable elements on a screenshot and describes what each one does, so a VLM can click and type without accessibility APIs. It reached 39.5% on the ScreenSpot-Pro grounding benchmark at release and powers the OmniTool Windows agent harness.
It is a perception component, not an agent — you pair it with a planner model, and grounding accuracy drops on dense or unusual UIs. The V1 line was MIT licensed; confirm V2's terms on the model page before use.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
- ScreenSpot-Pro GUI grounding: 39.5% (team-reported, Jan 2025)
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 7/10
Scored 7/10 (metrics tier, metrics-v1). Evidence: Microsoft Research, Feb 2025; checkpoints on HF under the microsoft org with OmniTool agent harness; Core perception layer for many open GUI-agent stacks in 2025; Perception-only component — must be paired with a planner VLM; grounding degrades on dense UIs; V1 was MIT; V2's exact license should be confirmed on the model page. License: V1 was MIT; confirm V2 terms on the Hugging Face model page before use.
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 7.0/10
- Maintenance · recency, releases, issue load
- 7.0/10
- Openness · license + docs
- 7.0/10
- Manual review · sanity read + red-flag check
- 7.0/10
- · Microsoft Research, Feb 2025; checkpoints on HF under the microsoft org with OmniTool agent harness
- · Core perception layer for many open GUI-agent stacks in 2025
- · Perception-only component — must be paired with a planner VLM; grounding degrades on dense UIs
- · V1 was MIT; V2's exact license should be confirmed on the model page