Flower Labs, a Cambridge University spinout, has launched a British frontier AI model that it says can match the performance of rivals from OpenAI and Anthropic.
The London and Hamburg-based start-up said its Endeavor general-purpose model, which draws on open-source software as well as its own proprietary intellectual property, completed tasks at similar levels to OpenAI’s GPT-5.6 Sol and Anthropic’s Claude Fable 5, two of the US companies’ main models.
Users of Endeavor can also deploy it on their local IT systems, which the company says gives them reliable access, control and improved security. The leading US providers with “closed-source” models require users to access them via software interfaces, access to which can be withdrawn.
“Europe should not have to rent its intelligence indefinitely from a handful of US companies,” said Professor Nicholas Lane, co-founder and chief scientist of Flower Labs and professor of machine-learning systems at the University of Cambridge.
“If AI is going to become fundamental infrastructure for companies and governments, then those organisations need a credible way to control the intelligence their systems depend on. Endeavor gives them that option without asking them to step back from the frontier.”
The launch comes as the government seeks to support UK-based alternatives to the big US AI companies in order to reduce the country’s dependence on overseas suppliers. On Monday it opened the first competitions under a £100 million sovereign AI research and development procurement scheme, which will run as a series of contests for AI start-ups. One goal, the government said, was to “significantly expand the UK’s public AI compute capacity while reducing costs” for researchers and businesses.
The UK’s reliance on leading US AI models was highlighted in June when the US government and Anthropic temporarily restricted access to the latter’s Fable 5 model over US national security concerns, prompting Downing Street to seek an exemption from the ban.
Lane, 47, and his co-founders Daniel Beutel and Taner Topal set up Flower Labs in 2023 as a spinout from Cambridge University. The same year they attended Y Combinator, the US start-up accelerator. They have since raised $23.6 million, and their models, AI agents and infrastructure have been used in more than 2,500 organisations, including the US bank JP Morgan, the IBM-owned software company Red Hat and the US Department of Energy.
Flower Labs’ technology draws on the capabilities of freely available AI models developed by companies including Meta and Mistral, the French artificial intelligence company, many of which perform well for specific tasks. It knits them together, adding the reasoning of its own model, called Lizzy, as well as proprietary harnesses, the software infrastructure that wraps around AI models to turn them into functional AI agents.
Lane, who teaches machine learning at St John’s College, Cambridge, said the difference in performance between closed-source models such as Anthropic’s Claude and open-weight models, systems where the final trained numerical parameters are placed in the public domain, had become “vanishingly narrow”.
“Open-weights are catching up to closed-source and if you are a strong, technical company you can look at ways of using the best of open-source to stand up these frontier models that we say are as good as the closed-sourced models,” Lane said.
He said Flower Labs’ long-term strategy was to develop more of its own proprietary models itself, and the company expects to raise additional capital in 12 months’ time.
Endeavor itself is a closed-source model, sold under licence. It enables organisations to use an AI training method called federated learning, which allows the model to be trained on sensitive data without that data being transferred to a central server. The company said this can help preserve privacy and security and is of interest to healthcare, financial services and defence companies as well as governments. Flower Labs has also developed its own AI deployment tools and services, for which it charges a fee.
The launch adds a British name to a field in which overseas challengers have already been positioning themselves as the alternative to Silicon Valley for organisations wary of handing over their data, with Canada’s Cohere tripling its London footprint to chase the same demand for so-called sovereign AI.
