Model comparison
ko-sroberta-multitask vs Gemini 2.5 Pro
Capability radar, value bars, and side-by-side posture.
ko-sroberta-multitask
Not scored
value score
Gemini 2.5 Pro
93
value score
ko-sroberta-multitask ctx
—
Gemini 2.5 Pro ctx
1.0M
Capability radar
Value · context · multimodal · openness · speed posture
Value head-to-head
- Gemini 2.5 Pro93
Attribute tape
| Field | ko-sroberta-multitask | Gemini 2.5 Pro |
|---|---|---|
| Developer | jhgan | Google DeepMind |
| Context | — | 1.0M |
| Modalities | Text, Code, Image, Audio, Video | |
| Openness | — | Proprietary Model |
| Speed | — | — |
| Price | — | $1.25/1M in · $10/1M out |
| Value score | — | 93 |
| Summary | A Korean sentence-embedding model built on RoBERTa and trained via multi-task learning on KorSTS and KorNLI datasets, producing 768-dimensional vectors for semantic search, text similarity, and clustering tasks. It is one of several ko-sentence-transformers variants, distinguished by its multitask training regime combining natural language inference and semantic textual similarity objectives. | Google frontier multimodal model. |