Frontier AI, under your control.
Endeavor 1.1
Frontier-class reasoning and coding, now with image understanding and an EU inference option. Use Endeavor through Flower or run it on infrastructure you control.
Frontier performance. Freedom to deploy.
GPQA
94.4
HumanEval
99.9
IFEval
94.5
AIME 2026
99.9
MMMU-Pro
81.1
Endeavor 1.1 is competitive with leading models from OpenAI and Anthropic, while giving you the choice to use it through Flower or deploy it on infrastructure you control. Build a lasting AI capability around Endeavor 1.1.
Frontier capability.
A foundation you can build on.
Endeavor 1.1 combines the performance of a leading frontier model with the freedom to choose where it runs, how you build on it and how you improve it around your work.
Competitive with the frontier
Endeavor 1.1 is evaluated directly against leading models from OpenAI and Anthropic, as well as the strongest open-source alternatives, including Kimi K3 and Nemotron 3 Ultra.
Across the five benchmarks, Endeavor 1.1 achieves the highest GPQA score (94.4) and HumanEval score (99.9), ties GPT-5.6 Sol and Claude Fable 5 on AIME 2026 (99.9), outperforms Kimi K3 on four of five benchmarks, and surpasses Nemotron 3 Ultra on all four text-based benchmarks. On MMMU-Pro, Endeavor 1.1 achieves a score of 81.1, remaining competitive with the other multimodal models. Nemotron 3 Ultra is excluded from this comparison because it is a text-only model.
The highest score in the comparison.
Matching the strongest OpenAI and Anthropic models evaluated.
Across benchmarks with available results.
| Benchmark | Endeavor 1.1 | GPT-5.6 Sol | Claude Fable 5 | Kimi K3 | Nemotron 3 Ultra |
|---|---|---|---|---|---|
| GPQA | 94.4 | 94.1 | 92.6 | 93.5 | 86.7 |
| HumanEval | 99.9 | 95.1 | 97.0 | 96.3 | 96.3 |
| IFEval | 94.5 | 95.9 | 91.7 | 92.8 | 91.9 |
| AIME 2026 | 99.9 | 99.9 | 99.9 | 96.7 | 94.2 |
| MMMU-Pro | 81.1 | 83.0 | 81.2 | 81.6 | N/A |
A generalist built for real work
Endeavor 1.1 is designed to act as the core generalist model across products, agents, and complex workflows, rather than being narrowly optimised for a single benchmark family.
Switch to Endeavor 1.1 today
Choose a production path today without giving up control over where Endeavor runs tomorrow.
Managed service
Global and EU GPU inference options.
Use Flower’s global offering or choose model inference running on EU GPUs. Click Request Access and tell us which inference location you need. Flower manages deployment, scaling, and model operations for you.
Private deployment
Keep the model and workloads in your environment.
Self-host Endeavor 1.1 in your own environment, with Flower support across your data, infrastructure, and systems.
Build on Endeavor 1.1 for the long term
How it is built
Built from a different starting point
Flower has spent years building the model and distributed systems technology required for powerful AI that organisations can trust, deploy on infrastructure they choose and improve around their own work. Endeavor 1.1 brings that foundation to the frontier.
Endeavor 1.0 followed Lizzy, a sovereign 7B model built for UK use. Endeavor 1.1 is our third model release this year, continuing Flower’s model programme with image understanding and an EU inference option.
From Lizzy to Endeavor 1.1
Lizzy
Sovereign 7B modelBuilt for UK use
Endeavor 1.0
Frontier generalistReasoning, coding and agents
Endeavor 1.1
MultimodalStronger capabilities across the boardEU inference option
The whole system matters
More than model weights
A modern frontier model is more than its weights. We optimise how Endeavor 1.1 reasons at inference time, how it builds and maintains context, how it uses tools, and how it checks, revises and recovers when a step fails.
We develop and evaluate these behaviours across multiple custom coding and agent harnesses, varying tool sets, interaction formats and execution environments so that performance transfers across real workflows rather than depending on a single benchmark setup.
Endeavor 1.1 system
Signals from real work
Evaluated where the work happens
Endeavor 1.1 is measured against today’s leading benchmarks, but it is not developed around them alone. FlowerBench gives us a different kind of signal by enabling repeatable evaluation on real, high-value enterprise workflows without moving the underlying proprietary data.
Tasks are contributed by organisations in the opt-in Flower Enterprise Evaluation Network and run inside their own environments. These end-to-end evaluations expose failure modes that public benchmarks rarely capture. We use those signals to guide Endeavor 1.1’s evaluation design, harness development, post-training and system-level improvements.
Open foundations, Flower advances
Build on what works. Advance what matters.
Endeavor 1.1 leverages mature, widely available capabilities from leading open-weight models, including general language understanding, public knowledge and common coding patterns. This lets us build on what the wider ecosystem already does well rather than recreating it from the ground up.
To this broad foundation, we add capabilities developed through Flower’s own model programme, including UK-specialist knowledge and reasoning from Lizzy, together with new model behaviours, specialist capabilities and training advances developed for Endeavor 1.1.
Continual pre-training, targeted post-training and careful model integration bring these strengths together and enable us to keep refining Endeavor 1.1 over time.
Start with Endeavor 1.1
Build intelligence that becomes increasingly your own.
Use Endeavor 1.1 through Flower today, then extend it with your agents, evals, data, and improvement loops. Turn frontier capability into a lasting AI foundation for your organisation.