South Africa-founded startup launches AI model with 10 million-token memory

Refiant AI, a South Africa-founded startup that uses algorithms to compress artificial intelligence (AI) models, has launched Protea, a suite of large language models (LLMs) that the company says is capable of processing up to 10 million tokens in a single prompt. 

The models can process and retain significantly more information at once before generating a response, according to the company. Protea comes in three versions with context windows of one million, five million, and 10 million tokens and can be accessed for free through Refiant’s platform without a waitlist or approval process. 

The launch comes three months after Refiant AI raised a $5 million seed round led by VoLo Earth Ventures to expand its AI optimisation platform and deepen research partnerships. Protea marks Refiant’s first major product launch since its fundraising.

“Customers don’t need more waitlists,” Mathew Haswell, Defiant AI’s cofounder, said. “They need models they can test, break, and build with. Protea is live, and we want people to use it from day one.”  

Founded in 2025 by Haswell, Viroshan Naicker, and Siddharth Gutta, Refiant AI is building machine learning systems to reduce compute costs and improve AI model efficiency.

According to the company, at its maximum capacity, Protea can hold roughly 7.5 million words, allowing users to analyse large volumes of information without splitting it into smaller chunks. It added that such a level of context could allow legal teams to review hundreds of contracts in a single pass, insurers to analyse years of claims data, and engineering teams to process entire software codebases.

The launch comes as AI companies increasingly compete to expand the amount of information their models can process at once. Anthropic’s Claude supports context windows of up to 500,000 tokens on certain Enterprise offerings, while Google’s Gemini offers up to one million tokens on its higher-tier plans. 

As AI developers move to improve performance by expanding context windows, Refiant is betting that a larger AI memory will give enterprises an edge when analysing large datasets.

“Long-context AI has been talked about for over a year now, but hasn’t really been commercially available,” said Naicker, CEO and co-founder of Refiant. 

The company said it has already demonstrated an internal prototype capable of processing up to 100 million tokens and is exploring how to benchmark and bring the technology into production. It added that Protea is the first of three planned product releases.

True scale demands moving beyond surface-level integrations to robust execution. We’ve filtered the noise out of Moonshot 2026, optimising the conference strictly for high-calibre connections between startup founders, global financial operators, enterprise leaders and individuals rewiring Africa’s technical frameworks. Get 20% off Early Bird tickets for a limited time.