Reka Core: Our Frontier Class Multimodal Language Model — Reka AI

🌈 Abstract

The article introduces Reka Core, Reka's largest and most capable multimodal language model to date. It highlights Reka Core's performance, capabilities, and deployment flexibility, as well as Reka's partnerships with leading technology platforms and organizations.

🙋 Q&A

[01] Performance Highlights

1. How does Reka Core compare to other leading models in the industry?

• Reka Core is competitive with models from OpenAI, Anthropic, and Google across key industry-accepted evaluation metrics.
• On a total cost of ownership basis, Reka Core delivers outsized value given its capabilities and deployment flexibility.
• Reka Core is comparable to GPT-4V on MMMU, outperforms Claude-3 Opus on multimodal human evaluation, and surpasses Gemini Ultra on video tasks.
• On language tasks, Reka Core is competitive with other frontier models on well-established benchmarks.

2. What are the key capabilities of Reka Core?

• Multimodal (image and video) understanding
• 128K context window for ingesting and recalling large amounts of information
• Superb reasoning abilities, including language and math
• Top-tier code generation capabilities
• Multilingual fluency in 32 languages

[02] Deployment Flexibility and Partnerships

1. How is Reka Core made available to customers and partners?

• Reka Core, like Reka's other models, is available via API, on-premises, or on-device to satisfy the deployment constraints of customers and partners.

2. What are some of Reka's key partnerships?

• Reka has partnered with leading global technology platforms such as Snowflake, Oracle, and AI Singapore.
• These partnerships enable Reka's customers, organizations, and individuals around the world to benefit from and build with Reka models by democratizing access to multimodal technology.

3. What are the goals and benefits of Reka's partnerships?

• The partnerships aim to bring Reka Core's impressive industry-leading performance to customers through the partners' platforms and services.
• The partnerships allow enterprises to unlock more value from their data, streamline complex tasks, and realize cost efficiencies through Reka's dynamic multimodal and multilingual models.
• For AI Singapore, the partnership aims to develop highly capable open-source models for Southeast Asia as part of their goal to anchor deep national capabilities in AI.