Coher Aya VISION AI is launched with 23 languages

Canadian artificial intelligence starts COHERE was launched in 2019 Specifically targeting the institution, but independent research has shown that it was so far Struggle To get a lot of market share between third -party developers Compared to competing American models Like Openai and Anthropic, not to mention The rise of the Chinese opponent open source Deepseek.

However, COHERE continues to enhance its offers: Today, the Cohere Research Department for AI It announced the issuance of the first vision model, Aya VISIONA new open -weight multimedia model that merges language and vision capabilities and is characterized by chaos in supporting inputs in 23 different languages ​​that COHERE says in an official blog publication is “Half of the World Population”, which makes it like a wide international audience.

Aya Vision is designed to enhance the ability of artificial intelligence to interpret images, generate text and translate visual content into a natural language, making Amnesty International Multifunction easier and effective. This will be particularly useful for institutions and institutions working in multiple markets all over the world with different language preferences.

It is now available on the COHERE website on the web and on artificial intelligence symbols Embroidery and Kaggle Under a Creative Communication Support-Noncommercial 4.0 International (CC BY-NC 4.0)Allowing researchers and developers to use, modify and share a model for non -commercial purposes as long as the appropriate support is given.

Moreover, Aya Vision is available through WhatsAppAllow users interact with the form directly in a familiar environment.

This limits its use of institutions and as a move of paid applications or the workflow of making money, unfortunately.

It comes in 8 billion and 32 billion teacher publications (Teachers indicate the number of internal settings in the artificial intelligence model, including its weights and prejudices, with a more powerful and performance model).

Supports 23 languages ​​and count

Although the leadership of artificial intelligence models from its competitors can understand the text through multiple languages, the expansion of this ability to vision -based tasks is a challenge.

But Aya Vision overcomes this by allowing users to create photo comments, answer visual questions, translate images and perform the text -based language tasks in a variety of languages:

1. English language

2. French

3. German

4. Spanish

5. Italian

6. Portuguese

7. Japanese

8. Korean

9. Chinese

10. Arabic

11. Greek

12. Persian

13. Polish

14. Indonesian

15. Czech

16. Hebrew

17. Indian

18. Dutch

19. Romanian

20. Russian

21. Turkish

22. Ukrainian

23. Vietnamese

In the blog post, COHERE showed how Aya VISION is analyzing pictures and text on product packages and providing translations or interpretations. It can also determine and describe art patterns of different cultures, helping users to identify things and traditions through the optical understanding of the behalf.

Aya Vision capabilities have wide effects across multiple areas:

• Learn language and education: Users can translate images and describe them in multiple languages, which makes educational content easier.

• Preserving culture: The model can generate detailed descriptions of art, monuments and historical artifacts, and support cultural documents in the languages ​​active.

• Access tools: Ins artistic intelligence can help users who are visually impaired by providing detailed images in their mother -in -law.

• Global communication: Real -time multimedia translation allows organizations and individuals to communicate through languages ​​more effectively.

Strong performance and high efficiency through pioneering standards

One of Aya VISION’s prominent features is its efficiency and performance in relation to the size of the model. Although it is much smaller than some of the leading multimedia models, Aya Vision outperformed many more alternatives in many major standards.

• Aya VISION 8B surpasses Lama 90B, which is 11 times larger.

• Aya VISION 32B surpasses QWEN 72B, Llama 90B and Molmo 72B, all at least twice their size (or more).

• Standard results appear on AYAVISIONBENCH and M-WEDVISION AIA VISION 8B, winning rates of 79 %, and Aya Vision 32B up to 72 % of winning rates in the tasks of understanding multi-language images.

The visual comparison of efficiency for performance highlights the Aya Vision feature. As shown in the barter graph for efficiency for performance, Aya Vision 8B and 32B shows a better performance in its class in relation to the size of the teacher, outperforming much larger models while maintaining mathematical efficiency.

Technical innovations that operate Aya Vision

Cohere for Ai is due to the performance of Aya VISION to many major innovations:

• Synthetic illustrations: The model benefits from the generation of artificial data to enhance training on multimedia tasks.

• Simling multi -language data: By translating and reformulating data through languages, the model acquires a broader understanding of multi -language contexts.

• Merging the multimedia form: Advanced techniques combine visions from both vision and language models, and improve general performance.

These developments allow Aya Vision to process images and text more precisely while maintaining strong multi -language capabilities.

The performance improvement scheme offered step by step how additional innovations, including artificial control (SFT), models, and expansion, contribute to high winning rates in Aya Vision.

The effects of decision makers of the institution

Despite the apparent meeting of Aya Vision’s needs for the institution, companies may have difficulty benefiting from them a lot due to the conditions of non -commercially commercial licensing.

However, CEOs, CTOS, IT leaders and researchers, Amnesty International, may use models to explore multi-language and multi-mediated capabilities that AI drives inside their organizations-especially in research, preliminary models and standards.

Institutions can still use internal research and development, evaluate multi -language artificial intelligence and experience multimedia applications.

You will find CTOS teams and Aya Vision teams of highly efficient and open value that exceeds much larger alternatives with lower calculations request.

This makes it a useful tool for measuring against royal models, exploring the potential solutions that AI drives, and testing multimedia interactions before adhering to the commercial publishing strategy.

For data scientists and artificial intelligence researchers, Aya Vision is more useful.

Its open nature and strict standards provide a transparent basis for studying the behavior of the model, controlling non -commercial settings, and contributing to the opening of smart developments.

Whether it is used for internal research, academic cooperation, or artificial intelligence ethics assessments, Aya Vision is an advanced resource for institutions that look forward to staying at the forefront of Amnesty International-Media-without restrictions on closed ownership models.

Open source research and cooperation

Aya VISION is part of Aya, and it is a wider initiative through COHERE that focuses on making artificial intelligence and relevant technology more multi -language.

since Beginning in February 2024The AIA initiative participated in a global research community that includes more than 3000 independent researchers in 119 countries, working together to improve the AI ​​language models.

To enhance its commitment to open science, COHERE has released the open weights of both Aya VISION 8B and 32B on the Kagge and Ungging face, ensuring that researchers around the world can reach models and experience models. In addition, Coher for Ai AyavisionBenchmark, a new multi -language assessment set designed to provide a strict evaluation framework for a multimedia AI.

The availability of Aya Vision as an open -weight model is an important step in making AI’s multi -language research more comprehensive and accessible.

Aya Vision depends on success Ayat EXPANSEAnother LLM family from Cohere for AI focuses on a multi -language AI. By expanding its focus on multimedia intelligence, Cohere for AI puts Aya’s vision as a major tool for researchers, developers and companies looking to integrate multi -language artificial intelligence into the workflow.

As the AIA initiative continues to develop, Cohere for AI has also announced plans to launch a new cooperative research in the coming weeks. Researchers and developers interested in contributing to the progress of multi -language artificial intelligence can join the open science community or apply for research grants.

Currently, Aya Vision’s launch represents a big jump in Multimbiamental Amnesty International, providing a highly performance and open solution that challenges the dominance of the largest and closed models. By providing these developments to the broader research community, Cohere for Ai continues to pay the limits of what is possible in multi -language communications that depend on artificial intelligence.