Image: Flower Labs
Flower Labs has released Endeavor 1.1, a new AI model that the company says can compete with the best models from OpenAI and Anthropic. Flower announced the model on November 3 and shared it on X, calling it a huge leap over Endeavor 1.0.
Introducing Endeavor 1.1.
— Flower (@flwrlabs) November 3, 2026
Intelligence to rival leading @OpenAI and @AnthropicAI models. Under your control.
A huge leap from Endeavor 1.0 to 1.1. Better reasoning, coding and agentic work. Multimodal is here. New EU GPU inference options with enterprise support.
Our third model release this year. Request access via link in thread.
Flower Labs is best known for Flower, a platform for training AI across many computers without moving the data to one place. Endeavor 1.1 is its third model this year. It follows Lizzy, a small model built for UK use that came out in April, and Endeavor 1.0, which arrived in September.
The main pitch is control. Flower says many companies today must pick between a powerful model they rent through another company's service and a weaker model they can run themselves. Endeavor, it says, offers both. Customers can use it through a service Flower runs or install it on their own servers or private cloud.
The new version adds image understanding, so it can read charts, screenshots and diagrams along with text. Flower also says it is stronger at reasoning, coding and agent work, meaning long jobs where the AI carries out many steps on its own using files, code and other tools.
Flower backs its claims with benchmark results, which are standard tests used to compare AI models. On a combined score that gives equal weight to four tests in science, math, coding and following instructions, Flower says Endeavor 1.1 scored 97.18. It lists GPT-5.6 Sol at 96.25, Claude Fable 5 at 95.30, Kimi K3 at 94.83 and Nemotron 3 Ultra at 92.28.
Image: Flower Labs
According to Flower's model page, Endeavor 1.1 had the top score on the GPQA science test at 94.4 and on the HumanEval coding test at 99.9. It tied GPT-5.6 Sol and Claude Fable 5 at 99.9 on the AIME 2026 math test. It did trail GPT-5.6 Sol on the IFEval instruction test, and on the MMMU-Pro image test it scored 81.1, behind GPT-5.6 Sol at 83.0.
Image: Flower Labs
Compared with Endeavor 1.0, the GPQA score rose from 92.0 to 94.4 and HumanEval went from 98.2 to 99.9. Flower says the GPQA gain means about 30% fewer errors on that test.
Image: Flower Labs
All of these numbers come from Flower's own testing. The company has not released the model for anyone to download, and outside groups had not published their own results at the time of this writing.
Another selling point is where the model runs. Starting with this version, customers can choose to have Endeavor run on graphics chips located in the European Union, which can matter for companies that must keep data inside the region. Flower says it plans to add UK only options and regional choices in the Middle East, Asia and Africa.
Flower also says customers can read the model's full thinking traces, the step by step notes a model writes while working through a problem. The company argues that OpenAI shows only summaries and Anthropic usually holds raw traces back. Flower adds that customers own what the model produces and may use it to train their own models.
On how it was built, Flower says Endeavor 1.1 starts from abilities found in leading open models and adds training and features from its own work, including Lizzy.
Businesses can request access through Flower's website. The company has not shared pricing publicly.
Comments