What is Flower Labs Endeavor 1.1?

Endeavor 1.1 is Flower Labs' flagship frontier-class language model, launched on November 3, 2026. It supports image inputs alongside text, delivers frontier-tier benchmark scores (GPQA 94.4, HumanEval 99.9), and adds an EU inference option that routes all processing to GPUs located inside the European Union.

How does Endeavor 1.1 compare to GPT-6 and Claude Opus 5.5?

Flower Labs positions Endeavor 1.1 as competitive with leading offerings from OpenAI and Anthropic on reasoning and coding tasks. On GPQA it scores 94.4 (up from 92.0 in 1.0) and on HumanEval it scores 99.9 (up from 98.2). The key differentiator is not raw benchmark rank but sovereign deployment: Endeavor can run entirely within your own infrastructure or on EU-region GPUs, which no closed frontier API guarantees.

What deployment options does Endeavor 1.1 offer?

Three options: Flower-managed global service, Flower-managed EU-only service (inference on GPUs located inside the EU), and private deployment in your own cloud or on-premises environment. All three share the same model weights and come with optional SLAs and Flower engineering support.

Can enterprises train custom models on Endeavor outputs?

Yes. Flower explicitly grants enterprises ownership of their inference outputs and thinking traces. You can use those outputs to train your own custom models, build evaluation datasets, and create fine-tuned variants, all without surrendering IP to a third-party provider.

Why does EU-only inference matter under GDPR and the EU AI Act?

Under GDPR, sending personal data to a model API that routes traffic outside the EU creates transfer and controller obligations that are difficult to satisfy with standard DPA clauses alone. EU-only inference eliminates the cross-border transfer issue for most processing scenarios. Under the EU AI Act, enterprises deploying high-risk AI applications also need robust audit trails and technical documentation; Endeavor's readable thinking traces support both requirements.

Flower Labs Endeavor 1.1: Frontier AI With EU Data Control

On November 3, 2026, Flower Labs launched Endeavor 1.1, the third model in its 2026 programme and its first with EU-region inference and native image understanding. The release arrives as enterprise AI teams in Europe and regulated industries worldwide face a genuine constraint: the most capable models are available only through US-hosted APIs, with no guaranteed data residency. Endeavor 1.1 is a direct response to that constraint.

What Is New in Endeavor 1.1

Flower Labs shipped three meaningful upgrades over Endeavor 1.0, which launched in September 2026.

EU inference. Starting with 1.1, customers can route all inference to GPUs physically located inside the European Union. Flower runs the service under SLA, so teams get frontier capability with no cross-border data transfer. The company also plans UK-only, Middle East, Asian, and African regions in future releases.

Multimodal input. Endeavor 1.1 accepts images alongside text. Flower represents visual signals as discrete tokens in the same backbone as text, so the model handles vision and language without a separate pipeline. This matters for agentic coding workflows where screenshots, UI flows, and architecture diagrams are part of the task context.

Stronger reasoning and coding. On the graduate-level science reasoning benchmark GPQA, Endeavor 1.1 scores 94.4, up from 92.0 in Endeavor 1.0. On HumanEval, it scores 99.9, up from 98.2. The gains come from targeted post-training on long-horizon agentic trajectories and what Flower describes as FlowerBench, an evaluation suite built from real enterprise workflows contributed by organizations in its opt-in enterprise evaluation network.

Benchmark Comparison

Model	GPQA	HumanEval	Context	Sovereign Deployment
Endeavor 1.1	94.4	99.9	128K (1M max)	Yes (EU, on-prem, private cloud)
Endeavor 1.0	92.0	98.2	128K	Managed only
GPT-6 / Astra	Competitive	Competitive	Varies	No EU guarantee
Claude Opus 5.5	Competitive	Competitive	200K	No EU guarantee

Benchmark figures from Flower Labs official documentation for Endeavor versions. Scores for GPT-6 and Claude Opus 5.5 are broadly competitive at the frontier tier but are not directly comparable without a single-table evaluation.

The EU Inference Problem Flower Is Solving

Flower’s blog post is pointed on the market gap. Azure’s UK-only per-token inference offering is limited to GPT-4o and what Flower calls an now 18-month-old GPT-4.1 mini model. The gap between “globally available frontier model” and “regionally constrained frontier model” is wide, and for enterprises operating under GDPR, financial services regulation, or the EU AI Act, that gap has real legal weight.

Sending sensitive prompts to a model that may route processing through any available data center globally creates GDPR Article 46 transfer obligations that standard Data Processing Agreements do not always satisfy for inference workloads. The issue is especially acute for use cases like document review, medical coding, financial analysis, and HR automation, all categories where prompt content regularly contains personal or regulated data.

EU-only inference removes the cross-border transfer argument for most processing scenarios. It does not solve all compliance requirements, but it removes one of the most common blockers to production deployment of frontier AI in regulated European organizations.

Three Deployment Paths

Endeavor 1.1 ships with three deployment configurations, each addressing a different enterprise risk profile.

Flower-managed global service. The standard API path, suitable for teams that want frontier capability quickly and accept a US-based provider with standard DPAs. Model ID is flwrlabs/endeavor-1.1; context window is 128K tokens in the supplied Codex configuration with automatic compaction at 96K.

Flower-managed EU service. Inference runs on GPUs located inside the EU. Same model, same API, same SLAs, different routing. This is the option most GDPR-conscious teams will start with for production workloads involving personal data.

Private deployment. You run Endeavor on your own infrastructure, with Flower providing deployment support, model artifacts (via private distribution confirmed during onboarding), and ongoing engineering assistance. You choose when to take updates. Your core AI capability does not depend on a provider’s continued availability or pricing decisions.

The private deployment path is particularly significant for enterprises building proprietary evaluation datasets, fine-tuning pipelines, or agent loops around a specific model version. With closed APIs, a model deprecation can break a year of investment in evals, prompts, and agent harnesses. With a model you control, that investment survives provider changes.

What Enterprises Can Do With Outputs

One underreported feature of the Endeavor programme is output ownership. Flower grants enterprises ownership of their inference outputs and the readable thinking traces Endeavor produces. You can use those outputs to build internal evaluation sets, create fine-tuned variants, and train downstream models, all without licensing restrictions.

This matters for AI-native teams that want a compound improvement loop: the more work the model does for you, the more training signal you accumulate, and the stronger your custom capabilities become over time. With a standard frontier API, that loop either does not run or it runs with the provider as the primary beneficiary of the signal.

For enterprise teams building toward what Enera describes as AI-native transformation, where AI is not a tool bolt-on but a core operational capability, the ability to accumulate and reuse inference data is a meaningful structural advantage over teams that run exclusively on closed APIs.

Governance and Multi-Agent Implications

Endeavor 1.1 also ships with readable thinking traces, meaning the model’s reasoning steps are inspectable before and after a decision. For multi-agent systems where one agent orchestrates several others, readable traces make it possible to audit which agent made which decision and why. This is one of the more important unresolved problems in enterprise multi-agent governance, and Flower’s implementation addresses it at the model level rather than requiring a separate observability layer.

The EuroAgent Stack, which Flower demonstrated at the European Frontier AI event in Berlin on October 13, 2026, showed how Endeavor integrates with federated agent architectures that keep sensitive data on separate organizational infrastructure while still collaborating across boundaries. The ARIC Hamburg announcement described it as combining European frontier models, collaborative agents, and sovereign infrastructure into a single stack.

What Enterprise AI Teams Should Do Now

Endeavor 1.1 access is available by request, with Flower onboarding organizations and partners on a rolling basis. Teams interested in testing it should request access through flower.ai and evaluate it on the specific tasks that matter to their production workloads rather than on generic public benchmarks.

The model represents a serious option for three categories of enterprise:

Regulated European organizations where GDPR or sector-specific rules make US-hosted API inference operationally or legally difficult.
Teams building proprietary AI capabilities that want to own their model outputs and training data rather than contributing signal to a provider.
Enterprises with stringent uptime or change-management requirements, where private deployment and controlled upgrade cycles matter more than access to the absolute latest model version.

If your team is ready to move beyond the limitations of US-only frontier APIs, Enera can help you build an AI architecture designed for long-term enterprise control.