UI-TARS 2

Name: UI-TARS 2
Author: ByteDance

ByteDanceProprietaryPending Human Review

UI-TARS 2 is ByteDance’s native GUI-centered research model, first published in its September 2 technical report. It starts from Seed1.6 checkpoints and combines a 230B-parameter mixture-of-experts language backbone, activating 23B parameters, with a 532M vision encoder. Multi-turn reinforcement learning trains language reasoning and actions across GUI, game, terminal, file-system, and tool environments. Parameter interpolation combines specialized agents into a broader model. The published research and demonstration qualify under the historical research-model rule, but reviewed sources do not establish publicly downloadable UI-TARS 2 weights or a general public model API.

2025-09-02

230B language backbone, 23B active; 532M vision encoder

Multimodal Mixture-of-Experts Transformer with vision encoder

Undisclosed; research model

Compare with other models

Specifications

Parameters: 230B language backbone, 23B active; 532M vision encoder
Architecture: Multimodal Mixture-of-Experts Transformer with vision encoder
License: Undisclosed; research model
Type: multimodal
Modalities: textimage

Benchmark Scores

Advanced Specifications

Model Family: UI-TARS
Finetuned From: Seed1.6
API Access: Not Available
Chat Interface: Not Available

Capabilities & Limitations

Capabilities: GUI reasoningtool usecodinggame interactionlanguage reasoning
Known Limitations: Public weights were not verifiedResearch results depend on the unified sandbox and agent harnessLong trajectories remain vulnerable to execution and planning errors
Notable Use Cases: general computer-agent researchmulti-turn reinforcement-learning researchterminal and GUI workflow evaluation
Tool Use Support: Yes

Resources

Related Models

SeedRealtime

ByteDance

SeedRealtime is ByteDance’s native full-duplex audio-visual language model for continuous conversation. It unifies text, audio, video, conversational timing, and response generation in one architecture. Streaming perception allows it to use the user’s environment to resolve ambiguous speech and maintain context in noisy multi-speaker scenes. The model was publicly described as fully rolled out in ByteDance products. Future tool-driven actions remain roadmap work; the launch does not establish downloadable weights or a general developer API.

Doubao Seed 2.1 Pro

ByteDance

Doubao Seed 2.1 Pro is ByteDance’s flagship productivity model in the Seed 2.1 family. It combines language reasoning with image and video understanding for document-heavy tasks, visual analysis, and multi-step agents. Public inference is offered through Volcengine and ByteDance applications, with no downloadable weights. The tier is a separate product identity; API snapshots and later service migrations are variants. Parameter counts and unsupported endpoint limits are left unspecified.

Doubao Seed 2.1 Turbo

ByteDance

Doubao Seed 2.1 Turbo is ByteDance’s efficient productivity tier in the Seed 2.1 family. It combines language reasoning with image and video understanding for document-heavy tasks, visual analysis, and multi-step agents. Public inference is offered through Volcengine and ByteDance applications, with no downloadable weights. The tier is a separate product identity; API snapshots and later service migrations are variants. Parameter counts and unsupported endpoint limits are left unspecified.