South Africa-founded artificial intelligence startup Refiant AI has launched Protea, a new suite of large language models designed to process up to 10 million tokens in a single prompt. The launch places the young company among a growing group of AI developers seeking to push the boundaries of how much information models can understand and retain during a single interaction.<\/p>\n\n\n\n
Founded in 2025 by Mathew Haswell, Viroshan Naicker and Siddharth Gutta, Refiant AI is developing technologies focused on making AI models more efficient and reducing the computing resources required to operate them. The company recently raised $5 million in a seed funding round led by VoLo Earth Ventures, providing capital to expand its AI optimisation platform and research partnerships. <\/p>\n\n\n\n
Protea is available in three versions, offering context windows of one million, five million and 10 million tokens. The models are accessible through Refiant’s platform without a waitlist or approval process, allowing developers and businesses to test the technology directly.<\/p>\n\n\n\n
The significance of the launch lies in Protea’s ability to process extremely large volumes of information at once. According to the company, the 10-million-token version can hold the equivalent of roughly 7.5 million words in context. This could allow users to analyse extensive documents and datasets without repeatedly dividing them into smaller sections.<\/p>\n\n\n\n
The potential applications are wide-ranging. Legal professionals could use long-context AI to examine hundreds of contracts in a single workflow, while insurance companies could analyse years of claims information. Engineering teams could also process large software codebases, potentially making it easier to identify patterns, errors and relationships across complex projects.<\/p>\n\n\n\n
Long-context capabilities are becoming an increasingly important area of competition in the AI industry. Major technology companies are also expanding the amount of information their models can process, making context length a key differentiator for developers and enterprise users.<\/p>\n\n\n\n
Refiant AI’s approach is particularly notable because it comes from an African-founded startup operating in a highly competitive global market. The company’s launch demonstrates that African technology entrepreneurs are increasingly participating in advanced AI development rather than focusing exclusively on applications built on top of models created elsewhere.<\/p>\n\n\n\n
The startup is also looking beyond its current capabilities. It says it has developed an internal prototype capable of processing up to 100 million tokens and is exploring how to benchmark and eventually bring that technology into production. Protea is reportedly the first of three planned product releases, suggesting that the company intends to continue developing its long-context AI technology. <\/p>\n\n\n\n
The challenge will be turning impressive technical capabilities into practical products that businesses can use reliably and affordably. Processing larger context windows can require significant computing resources, and companies will need to demonstrate that the benefits justify the costs.<\/p>\n\n\n\n
Still, Protea’s launch represents an important development for Africa’s emerging AI ecosystem. By combining long-context capabilities with a focus on efficient AI models, Refiant AI is positioning itself in a rapidly evolving global market where the ability to understand larger volumes of information could become increasingly valuable.<\/p>\n\n\n\n