Google Research
CoCa (Capturing Compositionality)
Efficiently combines image understanding and language generation.
Efficient joint training of vision and lStrong performance on captioning and VQAScalable architectureFoundation for multimodal models
Today's score
85.0
Where it ranks today
Best for / Not great for
Best for
- Image captioning
- Visual Question Answering (VQA)
- General image-text understanding tasks
- Research in efficient multimodal learning
Not great for
- Audio/video generation
- Real-time interactive applications
- Creative text generation beyond descriptions
Why it ranks here
CoCa represented a significant step in efficient multimodal training. While not a standalone product, its architectural innovations and performance benchmarks continue to influence the development of larger, more capable multimodal systems, particularly in research settings.
30-day trend
Score breakdown
Search trends84
Benchmarks86
Developer buzz85
News mentions84
Pricing
API: $0.00 in · $0.00 out per 1M tokens · Consumer: $0.00/mo
Pricing plans
Popular
Research Code
Open-source implementation for research.
Free
- Model architecture
- Training code
- Dataset integration
- Academic focus