Reka Core 2.0 Review: The Multimodal Dark Horse
Reka Core 2.0 brings surprising multimodal capabilities at a fraction of the cost of leading models. Full review and benchmarks inside.
The Underdog Rises
Reka has been quietly building one of the most capable multimodal models on the market. Core 2.0, released in early 2026, processes text, images, video, and audio natively — making it one of only three models (alongside Gemini 3 and GPT-5.2) with full quadri-modal support.
What sets Reka apart is pricing: at $1.50/M input tokens and $6/M output tokens, it's significantly cheaper than comparable models while delivering surprisingly competitive performance.
Video & Image Understanding
Reka Core 2.0 scores 78.4% on our video understanding benchmark, compared to Gemini 3 Pro's 84.2% and GPT-5.2's 81.1%. For image analysis, it achieves 82.9% accuracy on complex scene description tasks.
The model handles up to 30 minutes of video input and can extract detailed scene-by-scene summaries, identify objects and actions, and answer questions about temporal sequences.
Text Performance
On pure text tasks, Reka Core 2.0 is competitive but not class-leading. It scores 88.1% on MMLU-Pro, compared to GPT-5.2's 94.2% and Claude 4.5's 92.8%. However, for many practical applications, this level of performance is more than sufficient.
The 64K context window is adequate for most use cases but falls short of the 128K+ offered by leading competitors.
Best Use Cases
Reka Core 2.0 is ideal for multimodal applications on a budget — content moderation, media analysis, accessibility tools, and entertainment applications where cost efficiency matters more than absolute peak performance.
Explore Reka Core 2.0 alongside other multimodal models on Vincony.com.