UI-TARS
https://github.com/bytedance/UI-TARS
ByteDance Seed's open-source GUI-agent foundation model — released Jan 2025; trained end-to-end on screenshot → action pairs. Comes with UI-TARS-Desktop (Electron app) that runs the model against the user's screen. SOTA on OSWorld at release; weights on HuggingFace.
At a glance
- Type
- OSS GUI-agent foundation model + desktop app (ByteDance)
- Tier
- T2
- Section
- Computer-use & desktop agents
- Created
- 2025-01
- Latest release
- UI-TARS-1.5 (Apr 2025)
- License
- Apache 2.0
- Pricing
- OSS weights free
- Funding
- ByteDance internal (Seed team)
Taxonomy
- storage
- kv
- retrieval
- injection
- persistence
- session
- update
- agent-controlled
- unit
- turn
- governance
- opaque
- conflict
- stateless
When to use
Optimised for: End-to-end GUI-agent foundation model — screenshot → action
Anti-fit: Research-grade — no SLA / hosted endpoint from ByteDance; 7B baseline is small
Pros & cons
Pros
First major open-weights GUI agent FM; SOTA at release; Apache 2.0; full training recipes.
Cons
ByteDance origin = some enterprise data-residency concerns; no hosted endpoint; 7B baseline.
Claims & capabilities
arXiv 2501.12326 (Jan 2025); OSWorld 24.6% beating Claude 3.5 Sonnet (22.0%); weights on HF
Technical surface
- API surface
- Python / VLLM; desktop app for end users
- Backend storage
- Caller-side; no built-in storage
- Deployment
- Self-host (HuggingFace weights) + UI-TARS-Desktop Electron app
- Embedding model
- not applicable — not a memory product
- Multi-tenancy
- Single-process per model instance
- MCP
- searched not found
- A2A
- searched not found
- OpenTelemetry
- searched not found
Similar systems
Other computer-use & desktop agents in the catalog, ranked by inbound references.
- Browser Use T1
Python library that turns any LLM into a browser agent — extracts DOM into a structured representation the LLM can act on (rather than pure screenshot). Browser Use Inc. raised $17M seed Q1 2025 (Felicis); cloud product at cloud.browser-use.com. Project is one of the most-starred browser-agent OSS repos (~60k stars).
- Bytebot T3
Open-source desktop-agent runtime that runs an LLM-controlled Linux desktop inside Docker — uses xdotool / X11 capture for control. Pitched as the 'self-hosted Operator' alternative; agent sees a real desktop screen, not just a browser.
- Claude for Chrome T1
Anthropic's Chrome extension that lets Claude act inside the user's existing browser (rather than a sandboxed one). Research preview launched August 2025 — limited to Claude Max tier. Distinct from Anthropic Computer Use (which is the API substrate); this is the consumer surface that uses it inside Chrome.
- Highlight AI T2
Always-on desktop AI assistant that watches the user's screen (across all apps) and answers questions / takes actions based on what's currently visible. Mac + Windows. Founded by Mac Bishop (ex-Mixpanel); raised $13M Series A from Aspect / Lightspeed (Oct 2024). Distinguishes itself from Rewind by being action-oriented (not just recall).
- Hyperbrowser T2
Cloud-hosted headless browser API tuned for AI agents — captcha solving, stealth mode, session recordings. Y Combinator S24; pitched as a faster / cheaper Browserbase alternative. Used as substrate by agents that need browser execution but don't want to run their own Chromium fleet.
- Lutra AI T2
AI agent platform that automates SaaS workflows across Gmail, Slack, Sheets, etc. — uses natural-language instructions that compile to durable 'playbooks.' Founded by ex-Google/Facebook engineers; raised seed funding 2024. Positions itself as Zapier-meets-Claude for knowledge workers.