Flower Endeavor 1.1 Adds EU Inference and Image Inputs with Full Data Ownership
Flower Labs launched Endeavor 1.1 with image support, EU inference, and the right to use its raw thinking traces to train your own models.
Models · Source: Flower AI
What happened
Flower Labs released Endeavor 1.1 on November 3, 2026. The new frontier model adds multimodal image support alongside text processing. It also brings stronger reasoning capabilities for demanding agent workflows. Customers can access the model through Flower-managed global APIs. They can also choose specific inference on GPUs located exclusively inside the European Union. Flower plans to expand these regional options to the UK, the Middle East, Asia, and Africa.
The release focuses heavily on deployment flexibility and data ownership. Companies can run the model on premises or in a private cloud to keep sensitive workloads secure. Operating a model safely takes substantial in-house expertise. Flower aims to bridge this gap by offering enterprise support and service-level agreements. Unlike OpenAI and Anthropic, Flower allows users to inspect readable thinking traces. Users can retain all generated outputs and traces to train their own competing or specialized models.
Flower reports significant performance gains over version 1.0. GPQA scores rose from 92.0 to 94.4, representing about thirty percent fewer errors. HumanEval scores increased from 98.2 to 99.9. The company claims these scores outperform leading offerings from OpenAI and Anthropic. This claim is based on a composite equally weighting four selected benchmarks. Access is currently by request only. Customers go through a technical onboarding process to define usage limits, supported interfaces, and model artifact distribution.
Key facts
- 94.4 — Endeavor 1.1 GPQA benchmark score reported by Flower
- 99.9 — Endeavor 1.1 HumanEval benchmark score reported by Flower
- 1,000,000 — Maximum context length in tokens listed in the model documentation
- 128,000 — Token limit in the supplied Codex client configuration
Why it matters
The major cloud providers treat non-US regions as an afterthought. Global APIs do not guarantee where your data gets processed. Azure offers a UK-only endpoint, but it is limited to older models like GPT-4o and GPT-4.1 mini. Flower is explicitly targeting European enterprises that need frontier AI but cannot send proprietary data to US servers. Offering guaranteed EU inference with strict service-level agreements gives compliance-heavy startups a viable path to production. You no longer have to choose between frontier intelligence and regulatory compliance.
The licensing terms change the fundamental economics of AI development. Incumbents restrict using their outputs to train alternative models. They also hide their raw thinking traces. By allowing builders to retain these traces and use them for training, Flower turns API spend into a reusable asset. You are no longer just renting intelligence from a massive vendor. You are building a proprietary dataset of verified fixes and reasoning steps. You can use this data to train smaller, cheaper specialist models for your own systems. Every reviewed example becomes intellectual property that stays within your company.
For builders
Train specialist models on raw traces
Flower lets you keep inference outputs and readable thinking traces. You can use these assets to fine-tune smaller models for recurring tasks without violating terms of service. You pay for the initial API calls, but you own the resulting intelligence. This is a massive opportunity for startups building specialized coding or troubleshooting agents.
Verify actual context limits
The model documentation claims a one million token maximum context length. However, effective request limits depend entirely on your deployment, upstream route, and internal processing overhead. The supplied Codex configuration only supports 128,000 tokens. Builders must verify their specific technical contract during onboarding before migrating large context workloads.
Deploy on premises for IP protection
Prompts and iterative usage can reveal proprietary code and product plans. Running Endeavor in your private cloud or on premises keeps sensitive workloads entirely within your environment. You control when to upgrade and avoid depending on a provider-hosted API. Enterprise customers pay for the infrastructure but gain total control over their intellectual property.
My take
OpenAI and Anthropic want to lock you into renting their brains forever. They hide their reasoning and restrict your right to train competing models. I love that Flower is calling their bluff by giving you the raw thinking traces and the legal right to build your own intelligence. If you are building enterprise AI in Europe, this is exactly the leverage you need to stop leaking intellectual property to American cloud providers.
Original reporting: Flower AI. This is my rewrite and opinion.