The UK's Sovereign AI Lab

Cosine launched Lumen, a coding-focused AI model family it calls the UK's sovereign AI lab, claiming its specialist models outperform generalist systems including GPT-5.4, Gemini 3.1 Pro, GPT-5.5, and Kimi K2.6 on its Niche-Bench benchmark. Cosine published an open letter on sovereign AI for the United Kingdom supported by a coalition of executives from critical national infrastructure, and offers Lumen in public cloud, managed single-tenant, and fully air-gapped deployments with post-training for COBOL, Fortran, Verilog, Rust, and complex SQL.

The Sovereign AI Lab We train AI models and specialised coding agents for organisations that need frontier capability inside secure environments. In coalition with Open Letter on Sovereign AI for the United Kingdom We have published an open letter, supported by a coalition of executives from critical national infrastructure. View Open Letter https://cosine.sh/sovereign-ai-open-letter Real benchmarks. Real results. Lumen is optimised for the outcomes that matter: shipping better code, faster, with fewer headaches. GPT-5.4 Gemini 3.1 Pro GPT-5.5 Kimi K2.6 Lumen Outpost Read more about Niche-Bench https://cosine.sh/blog/training-for-niche-languages-legacy-systems Our thesis: specialists outperform generalists Most AI models try to be everything to everyone. They write poetry, solve physics problems, and summarise novels. Lumen does one thing: write exceptional software. Our thesis is simple. A model trained exclusively on real production code, from real engineering teams will always beat a generalist at the task developers actually need. That's why every training decision, every optimisation, every evaluation metric is grounded in a single objective: better code. Less AI Slop Optimised for maintainability, readability, and production-quality outputs, not bloated code generation. Built for Real Systems Designed to reason across complex architectures, legacy infrastructure, and long-lived codebases. Beyond Mainstream Languages Post-trained for enterprise and niche languages including COBOL, Fortran, Verilog, Rust, and complex SQL. Use Cosine wherever you work Start in your terminal Work within a terminal-native interface, with a local-to-remote execution model, external tool access via MCP, and multi-agent orchestration. Collaborate in the Cloud Run multiple coding tasks in parallel, plan large changes and collaborate across engineers, PMs, and stakeholders in a shared project. Built to fit your security environment Start quickly in Cosine Cloud, or move into a dedicated tenant or fully air-gapped deployment when your infrastructure, data, and compliance requirements demand it. Public Cloud Fully managed Cosine for teams whose requirements fit our standard cloud controls. Managed single-tenant cloud A private Cosine environment operated for your organisation with enterprise controls. Fully air-gapped For organisations that need Cosine to run entirely inside their own security perimeter. Three models. One philosophy. Lumen isn't a single model — it's a system designed for every layer of the development workflow. Runs cheaply, operates on-device, and handles the heavy lifting around the coding loop.