South African artificial intelligence company Refiant has unveiled what it says is one of the world’s largest commercially available long-context AI models, marking a major step in the race to make generative AI more useful for complex enterprise applications.
The company on Wednesday launched Protea, a suite of AI models capable of processing up to 10 million tokens of information in a single prompt, far exceeding the context windows currently offered by most commercially available large language models.
The technology is immediately accessible through Refiant’s platform without waiting lists or approval requirements, a move the company believes will accelerate adoption among developers and businesses seeking to deploy AI on large-scale datasets.
Context windows determine how much information an AI model can process and retain during a single interaction. Existing models typically require large datasets to be broken into smaller segments before analysis, increasing processing time and reducing the model’s ability to understand relationships across documents.
Protea aims to remove that limitation by allowing organisations to analyse entire regulatory libraries, enterprise software codebases, historical financial records and extensive scientific datasets in a single pass.
According to Refiant, the 10-million-token context window is capable of holding approximately 7.5 million words or around 15,000 pages simultaneously, equivalent to several years of corporate communications or decades of archived reports for an individual user.
The expanded memory capacity is expected to unlock new enterprise applications across industries including financial services, healthcare, insurance and software engineering, where organisations increasingly require AI systems capable of understanding vast amounts of interconnected information.
Engineering teams, for example, could analyse complete software repositories without dividing projects into multiple files, while insurers could process years of claims histories in a single workflow.
Beyond increasing scale, the company says Protea addresses the long-standing “lost in the middle” problem, a technical limitation in long-context AI systems where models lose accuracy when reasoning over information buried deep within lengthy inputs.
Refiant was founded by Viroshan Naicker, Siddharth Gutta and Mathew Haswell, combining expertise spanning quantum mathematics, finance and commercial technology development.
The company’s underlying optimisation techniques draw inspiration from evolutionary search and swarm intelligence, approaches designed to improve the computational efficiency of large language models.
Before launching Protea, Refiant attracted industry attention after demonstrating techniques that compressed OpenAI’s GPT-OSS-120B model sufficiently to operate on a MacBook Pro equipped with 18GB of memory.
That work helped secure a $5 million seed funding round led by VoLo Earth Ventures and established research collaborations with Imperial College London and University College London’s Sargent Centre for Process Systems Engineering.
Viroshan Naicker, chief executive and co-founder said long-context AI has remained largely theoretical despite extensive discussion across the industry over the past year.
“Long-context AI has been talked about for over a year now, but hasn’t really been commercially available,” he said.
Mathew Haswell, chief product and operating officer said customers increasingly require production-ready tools rather than experimental demonstrations.
“Customers don’t need more waitlists. They need models they can test, break and build with. Protea is live, and we want people to use it from day one,” he said.
The launch represents the first stage of Refiant’s broader product roadmap. The company disclosed that it has already demonstrated an internal prototype with a 100-million-token context window and is evaluating pathways to commercial deployment as it continues developing next-generation enterprise AI systems.
