Seed1.5-VL
532M vision encoder plus 20B-active MoE LLM; SOTA on 38 of 60 public VLM benchmarks, ahead of OpenAI CUA on GUI control.
Technical report on building a compact vision-language model, with agent tasks such as GUI control and gameplay. Offered through Volcano Engine as doubao-1-5-thinking-vision-pro-250428.
- Date
- Sunday, 11 May 2025
- Lab
- ByteDance Seed
- Kind
- model
- Access
- closed API
Figures
| Measure | Value | Measured by |
|---|---|---|
| Public benchmarks with SOTA | 38 of 60 | company |
Authors' own evaluation.
Sources
This record was checked against its sources on 6 October 2026. How we check