REKA AI

Reka started as a brain drain from Big Tech. In 2023, five AI researchers walked out of Google DeepMind and Meta to start their own lab in San Francisco.

The standout is Yi Tay, who had co-led PaLM-2, one of Google's flagship language models, and invented research techniques other labs still use. Dani Yogatama, also ex-DeepMind, took the CEO seat.

Instead of joining the OpenAI gold rush, they bet they could build better models with a smaller and more senior team. In June 2023 they came out of stealth with $58 million already in the bank.

Reka builds AI models and sells access to them, mostly to businesses and governments. Companies pay to use Reka's models through an API or run them on their own servers.

The pitch is multimodal. One model that can read text, look at images, watch video, listen to audio, and parse a PDF.

Enterprises that do not want to hand their data to OpenAI are a natural customer. Basically, Reka sells a private, do-everything AI brain to companies that care about control.

Reka's models come in sizes. Reka Core is the flagship, a large multimodal model that handles text, images, video, and audio.

Reka Flash and Reka Edge are smaller, faster, cheaper versions for lighter jobs. Yasa-1 was the first model the company shipped, back in 2023.

The through-line is multimodality. Every Reka model is built to understand more than just text.

Reka's edge was never money, because it had far less of it than its rivals. The edge was talent and speed.

A small team of senior researchers shipped real multimodal models fast, starting with Yasa-1 in 2023. It leaned into enterprise and on-premise deployment, the part of the market OpenAI was slower to serve.

Strategic backers mattered too. Snowflake and Nvidia did not just invest, they gave Reka a route to enterprise customers and to chips.

The strategy is simple. Be smaller, be faster, be the AI you can actually run yourself.

Reka is fighting a war it cannot win on money. OpenAI, Anthropic, and Google can spend more on a single training run than Reka has raised in its entire life.

Frontier AI is brutally expensive, and the best researchers get poached constantly. For a lab this small, every advantage has to come from being smarter and faster, not richer.

Staying independent while the giants throw billions around is the real, ongoing test.

Reka pulled in serious names for such a young lab. Its 2023 seed round of $58 million was led by DST Global and Radical Ventures, with Snowflake Ventures joining and former GitHub CEO Nat Friedman investing as an angel.

Later, Nvidia and Snowflake helped push the company past a $1 billion valuation. When the biggest chipmaker and a major data company both back you, it is a bet that your models will run on their platforms.

Total funding sits around $110 million.