Refiant, an AI optimisation company, today launched Protea – its suite of long-context AI models – with a 10 million-token context window, one of the largest ever made publicly available, accessible immediately via refiant.ai.
Until now, AI has been constrained by how much it can see at once. The most capable models struggle with more than a few hundred thousand tokens in working memory, forcing elaborate workarounds to compensate for what the model can’t access.
Protea removes those constraints. Entire regulatory archives, enterprise codebases, decades of clinical trial data – datasets that previously had to be broken apart and fed to models in fragments – can now be processed in a single pass, with full fidelity.
The model is the first product of its kind to ship to production at this scale – delivering on what competitors have so far only promised. Users can start building today on models ranging from one to 10 million tokens, with no waitlist or approval process for free.
What 10 Million Represents
At 10 million tokens, Protea can hold approximately 7.5 million words in context or 15 000 pages back to back – enough to hold and understand up to 5 years worth of email or Slack messages or 20-30 years worth of documents or reports for a single person.
This opens the door to previously impossible enterprise use cases. Engineering teams could ingest an entire codebase and compress a month of analysis into a single day. Insurance firms could process years of claims data in one pass. For teams building agentic workflows, it enables agents to operate across vast context without losing track of earlier reasoning.
Protea successfully tackles the “lost in the middle” problem – a well-documented limitation of million-plus token windows, where models stay accurate at the start and end of the context but lose the thread on everything buried between.
The Refiant Approach
Refiant was founded by Dr Viroshan Naicker, Siddharth Gutta and Mathew Haswell – a team spanning quantum mathematics, traditional finance and commercial scaling – with a conviction that modern LLMs are fundamentally inefficient and that better approaches already exist in nature. Its methods draw on evolutionary search and swarm-style optimisation, mimicking how natural systems find efficient solutions to complex problems.
Refiant first applied these techniques to model compression, shrinking OpenAI’s GPT-OSS-120B to run on a MacBook Pro with 18GB of RAM. Those results helped secure a $5 million seed round led by VoLo Earth Ventures, as well as research partnerships with Imperial College London and UCL’s Sargent Centre for Process Systems Engineering.
Protea Is the Beginning
“Long-context AI has been talked about for over a year now, but hasn’t really been commercially available,” said Viroshan Naicker, CEO and co-founder of Refiant. According to Mathew Haswell CPO/COO co-founder.
“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.”
The Protea series is open and live, and Refiant is inviting teams to stress-test the context window across different industries and use cases. Internally, the company has already demonstrated a working prototype with a 100 million context window and is exploring how best to benchmark and productionise this at that scale in future.
The launch marks the first phase of a three-stage roadmap with further announcements to come over the next 3 months.
