New AI Model Reka Rises to Challenge ChatGPT, Claude and More
A startup specializing in developing custom AI models for large corporations has introduced Reka Core to the public, a multimodal language model capable of processing text, image, video, and audio inputs. Reka AI, founded in 2022 by researchers from Google’s DeepMind, Baidu, and Meta, has secured $60 million in funding from investors including New York Life Ventures, Radical Ventures, and DST Global.
Reka Core Versus Other AI Models
Reka Core is the company's most advanced model yet, and according to Reka AI, it can compete with larger and more well-funded models such as those from OpenAI, Anthropic, and Google. In various benchmarks, including the Massive Multi-discipline Multimodal Understanding and Reasoning (MMMU), Reka Core has shown strong performance, comparable to models like GPT-4V and outperforming others like Claude-3 Opus and Gemini Ultra on specific tasks.
Capabilities of Reka Core
Reka Core boasts multimodal capabilities for images and videos, strong reasoning abilities, code generation, and multilingual fluency. The model is accessible for free through a chatbot interface and is also available via API for developers, with pricing set at $10 for every 1 million tokens and $25 for the same number of output tokens.
Training and Data Usage
Reka Core was trained from scratch over several months using thousands of GPUs and textual data from 32 languages. The model has fluency in multiple languages, including English and various Asian and European languages. It also leveraged a multilingual training dataset from Wikipedia, enabling it to have a basic understanding of numerous uncommon languages and dialects.
Key Features of Reka Core
While Reka Core offers impressive visual analysis capabilities, it does not have the ability to generate images like other models such as ChatGPT Plus or Google Gemini. However, its vision capabilities are swift and accurate, making it suitable for tasks requiring visual interpretation.
Reka Core also serves as an effective coding assistant, although users may need to provide precise prompts for optimal results. The model comes with built-in safety controls to prevent the generation of harmful or unethical content, demonstrating a responsible approach to AI usage.
Reka Core's Creative Writing and Knowledge Capabilities
When it comes to creative writing, Reka Core produces clear and engaging narratives, although its contextual limitations may impact longer story generation. In terms of knowledge and reasoning, the model performs well, handling complex questions and offering logical explanations effectively.
Additionally, Reka Core excels in language comprehension, displaying the ability to understand text with errors, provide translations, and grasp nuances in different languages. The model has also demonstrated mathematical capabilities and clear logical reasoning.
Conclusion
Reka Core presents a compelling option for businesses seeking a versatile AI model with strong multimodal capabilities, language comprehension, and reasoning skills. While it may have limitations in certain areas compared to other models, its free accessibility and robust performance make it a noteworthy contender in the AI landscape.
