Reka AI, a Multimodal AI Startup, Raises $110M at $1 Billion Valuation

Reka AI Inc., an emerging multimodal AI research startup, announced today the completion of a $110 million funding round backed by industry giants Nvidia Corp. and Snowflake Inc.

Founded in 2022, the company specializes in developing multimodal models with optimized training and service infrastructure designed to deliver cost-effective inference solutions.

Its product portfolio includes Reka Flash - a flagship multimodal model capable of processing video, images, text and audio. The company recently expanded its offerings with Reka Vision and Reka Research, two widely accessible AI platforms that enable complex problem-solving and visual search capabilities.

According to Bloomberg reports, this funding round has propelled Reka's valuation to $1 billion, tripling from its previous $300 million valuation after a $50 million financing round involving Snowflake Ventures.

Last year, Snowflake reportedly expressed interest in acquiring Reka at three times its then-valuation. However, Bloomberg indicates the deal negotiations were terminated after both parties opted for independent development paths, though maintaining ongoing collaboration.

Vivek Raghunathan, Snowflake's Senior VP of AI Engineering, told Bloomberg that Reka will provide AI models and related services to Snowflake clients.

"There are very few teams capable of building what they've created," Raghunathan stated. "Almost all talent at this level is concentrated in companies like OpenAI, Meta or Anthropic. Reka stands out as one of the rare independent firms - they've proven their competitive capability."

The company recently launched Reka Flash 3.1, which demonstrates superior performance compared to similarly scaled models like Quen3-32B and OpenAI's o3-mini.

Reka Flash itself is a compact 21-billion-parameter model utilizing quantization techniques to reduce memory footprint, enabling efficient operation on mobile and edge devices with minimal resource requirements.

According to company claims, their Quant technology achieves near-lossless quantization of Reka Flash 3.1 with minimal performance degradation - only 1.6-point average decline in hard reasoning benchmarks, significantly better than the next best quantization method showing 6.7-point average decline.

Over the past year, Reka's workforce has grown from 20 to 50 employees. While initially focused on model optimization, the company has expanded into developing tools for enterprise AI integration and application logic implementation.

Reka plans to use the new funding to accelerate technical development of its models and expand its multimodal platform, aiming to drive broader enterprise adoption and global market expansion.