Flower Labs introduced Endeavor 1.0 on September 1, 2026, calling it a frontier-class generalist for long-horizon agent work that organizations can run as a managed service or deploy on infrastructure they control. Access starts as a preview. The company is onboarding a limited set of organizations and partners rather than opening self-service weights or APIs on day one.
Nicholas Lane, co-founder and chief scientific officer, wrote the post with researchers Lorenzo Sani and Yan Gao. Endeavor arrives four months after Lizzy, Flower’s 7 billion parameter sovereign model built for UK use. The lab says Endeavor is the next step: a broad model meant to compete with closed systems from OpenAI and Anthropic, while leaving a path to private deployment.
Scores, then a private preview
Flower posted four launch scores: 92.0 on GPQA, 98.2 on HumanEval, 99.9 on AIME 2026, and 94.1 on IFEval. On that short table, Endeavor leads the group on HumanEval, ties GPT-5.6 Sol and Claude Fable 5 on AIME 2026, and beats Kimi K3 on three of the four tests. The company said no small benchmark set proves usefulness, and pointed to FlowerBench, an evaluation network that runs opted-in enterprise tasks inside the customer’s own environment.
Endeavor can be called through familiar model APIs. Flower will operate a production service for teams that want the lab to handle scaling. The same model, it says, can move into a private environment later, or run mixed: managed for most work, private for sensitive jobs. The company said it builds on widely available open-weight strengths rather than retraining every general skill, then adds its own post-training, including UK-specialist work from Lizzy.
Decoded Take
Flower is selling an exit ramp from closed APIs at a moment when UK and EU buyers are being told to worry about dependency. The scores are competitive on a four-row card. They are also the easy part. A preview that onboards by request is not yet a model you can put in a procurement. The harder claim is that agents, evals, and data pipelines built on Endeavor can move from Flower’s service into a customer’s cellar without a rewrite. Watch how many named UK or European organizations leave the preview with a private deployment, whether FlowerBench results ever become public enough to compare, and whether Lizzy’s sovereign story survives once Endeavor is the product everyone actually wants.