UI-TARS

UI-TARS

Hosted AgentTrustedCOMING SOON

by ByteDance

ByteDance's UI-TARS vision-language agent reads raw pixels for UI understanding — no DOM or accessibility tree required. High accuracy on visually complex, canvas-heavy, and non-standard layouts where selector-based automation breaks. Available hosted or self-deployed from the open Apache-2.0 weights.

This runner is not yet wired to the live routing layer. API requests targeting it return 503 runner_unavailable until the integration ships.

Provider
ByteDance
Type
Hosted Agent
Added
Jan 2025
Interface
Task-level
Trust tier
trusted
Egress
Vendor egress (unverified)
Evidence
partial
Max session
30 min
Regions
US, APAC
Pricing
per run $0.48self-host weights free (Apache-2.0)
Compliance
ZDRSOC 2HIPAAPCI-DSS ROC
Traffic share
Success
Median duration
$0.48/run
List price
p95 first action
Not enough live samples for a trend yet.

Pricing

ModelNative priceNormalized $/runNotes
per run$0.48/run
per run$0.48
self-host weightsfree (Apache-2.0)

Capability matrix

Capabilities for this runner — extra/gap badges compare against the most common capability set for Hosted Agent runners in the catalog.

Login flows
MFA/2FA interactions
CAPTCHA handling
File download
File upload
Form filling
Structured extraction
Vision-based
DOM-based
Stealth/anti-bot
Residential proxies
Persistent sessions
Parallel sessions
Human-in-the-loop
Full evidence
Storage capture

Code sample

Pin this runner via routing: "bytedance-ui-tars"

curl -X POST "$BANANA_PEEL_BASE/api/v1/responses" \
  -H "Authorization: Bearer $BANANA_PEEL_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
  "model": "banana-peel",
  "input": "Extract the title from https://example.com",
  "routing": "bytedance-ui-tars"
}'

Command Palette

Search for a command to run...