Cohere’s launches Aya Vision AI with support for 23 languages

Canadian AI Startup Launched in 2019 Specifically targeting the facility, but independent studies showed that it has been so far arrogant Earn many market share among third party developers Compared to owner US model providers Epenai and anthropic, not to note Chinese open source rival DeepSeek’s rise.

Again, it continues to increase our proposals: Non-profit research department for AI today Announced the release of the first vision model, the Vision VisionA new open multimodal AI model connecting language and vision and 23 different languages, 23 different languages ​​supported in a different blog post of different languages, “the population of the world” is said to be “the population of the world”.

Aya Vision is designed to comment on EU’s images, create text and translate visual content in natural language and translate into a natural language, to make the visual content more accessible and effective. It would be especially useful for enterprises and organizations operating in many markets around the world with different language options.

Already available on the COONE website and AI code communities Hug face and Kaggle a Creative Commons Attribution-Non Commercial Period 4.0 International (CC by-NC 4.0) LicenseAllows researchers and developers to change and share the model, which allows you to change and share the model for non-commercial purposes.

In addition, Moon Sight is available via WhatsAppallows users to interact with the model in a direct familiar environment.

This limits the use of as an engine for businesses and paid applications or paid work streams.

It comes in 8 billion and 32 billion parameter versions (Settings, stronger and more performing model, including more, weight and biases, including the number of internal parameters.

Supports 23 languages ​​and counts

Although the leading AI models of competitors can understand the text in many languages, this ability is a problem with the vision-based positions.

However, the moon view users to create image labels, to answer visual questions, translate the images and perform text-based language tasks in different languages.

1. English

2. French

3. German

4. Spanish

5. Italian

6. Portuguese

7. Japanese

8. Korean

9. The Chinese

10. Arabic

11. Greek

12. Persian

13. Poland

14. Indonesia

15. Czech

16. Hebrew

17. Hindi

18. Dutch

19. Romania

20. Russian

21. Turkey

22. Ukraine

23. Vietnamese

In the blog post, the Bohere, how the view of the Moon can make an image and how the text can analyze and present translations or explanations. You can also identify and describe art styles from different cultures, can help users learn about objects and traditions through a strong visual understanding of AI.

Opportunities for the moon have a wide range of imagination between the field:

• Language Learning and Education: Users can translate and describe the pictures in many languages, the content of education is more accessible.

• Cultural Protection: The model can create detailed descriptions of art, signs and historical works that support cultural documents around the world.

• Accessibility tools: Visually-based AI, with providing detailed image descriptions in the native language can help users who are weak.

• Global communication: Real-time multimodal translation organizations and individuals allow you to communicate more effectively among languages.

Strong performance and high efficiency in leading criteria

One of the sustainable features of the Moon Vision is its efficiency and performance compared to its model size. Although it is much smaller than some leading multimodal models, the Vision has prevailed in several main criteria in several main criteria.

• Aya Vision 8B Outperforms Llama 90B, which is 11 times larger.

• Aya Vision 32B Outperforms Qwen 72b, Llama 90B and Molmo 72b, all at least twice as large (or more).

• In Ayavisionbench and M-WildVision 8B, M-Wildvion 8B in M-Wildvion 8B and 79% Vision 32B, 72% of the concept of multilingual image reaches 72% of the concept.

The visual comparison of effectiveness emphasizes the advantage of the moon. As shown in efficiency, the performance graphic, the Moon, Vision 8B and 32B, protecting the efficiency of calculation, depreciating the best levels of performance than their parameter size.

Technological innovations that increase the ability to see the moon

For AI attributes, COUNE MUNA GETS PERSONALS TO Several basic innovations:

• Synthetic annots: The model uses synthetic information generations to increase training in multimodal tasks.

• Multilingual data scale: By translating data between data and turning back, the model gets a wider concept of multilingual contexts.

• Multimodal model combination: Advanced techniques combine both visual and language models and general performance improvement.

These improvements allow the moon to process the image and text more accurately while maintaining strong multilingual opportunities.

Step-by-step performance schedule, synthetic delicate adjustment (SFT), model combinations and scale, show how increased updates in places where the moon contributes to high victories.

The results of enterprise decisions

According to the venture, it may have difficulty using the restrictive non-profit license of enterprises, as the facility brings.

Nevertheless, CEOS, CTOs, IT leaders and AI researchers, especially research, prototypes and benchmarking can use models to explore multilingual and multimodal capabilities.

Enterprises can still use internal research and development, assessing multi-speaking AI performance to practice with multimodal applications.

CTOs and AI teams will find the ability to see the moon, which is valuable as a highly effective model, which is higher than a higher alternative, which is higher alternatives when required.

This is a useful tool for assessing special models, explore potential EU-controlled solutions and test multi-language multimodal interactions to commercial placement strategy.

The moon vision is more useful for information scientists and AI researchers.

Learning open source nature and strict fruits, model behavior provides a transparent foundation to contribute to fine regulation and AI progress in non-commercial parameters.

The use of internal research, academic cooperation or use for AI ethics assessments, serves as a most advanced resource for enterprises that wants to stay in front of the Multilingual and Multimodal AI – owner, closed source models.

Open source research and cooperation

The meeting is part of the month, a wider initiative, a wider initiative, a wider initiative with more multilingual.

Since then It happened in February 2024The Monea Initiative was conducted by 119 countries in the 119 independent researchers of 119 countries and 119 countries working together to improve the AI ​​models.

To further increase the obligation on open science, the MONA meeting left open weights for Kaggle and 32b, Kaggle and the embrace, released the open weights for the world researchers and models. In addition, COLARE for AI has introduced a new multilingual vision assessment set designed to provide a serious assessment framework for Multimodal AI.

The existence of the Vision as an open model of the Aya is an important step to be more comprehensive and accessible to a multilingual AI research.

The moon is based on vision success The width of the moonAnother LLM family, which is somewhat much for the EU, is directed to the multilingual AI. By expanding the attention of Multimodal AI, Colare for the EU is placed as a basic tool for researchers, developers and enterprises, which want to integrate the Multilingual AI on work flows.

As the Aya continues to develop, Co. for the EU also announced its plans to start joint research efforts in the coming weeks. Researchers and developers interested in contributing to multilingual AI progress can join the open science community or apply to research grants.

So far, the release of the verse, a significant leap in multitating multimodal EU, the larger, high-performance, which dominates higher, indoor sources, is an important leap that offers a high-performance, open solution. By making these progress in the more research community, CO for AI continues to push the borders of those possible in the AI-managed multilingual communication.