Flower

Introducing Endeavor 1.1.

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 1.43 is out.

You can now stop a federated learning run after any completed round with a should_stop callback, keeping the current model and collected metrics.

This release also reduces Agent response latency, adds experimental support for creating and listing persistent Agents, and moves SuperLink's Fleet and Control APIs fully to HTTP.

Upgrading a custom strategy? If it relies on automatic error replies, you'll need to handle unavailable nodes and missing replies explicitly. The standard Strategy.start() loop handles unavailable nodes automatically.

Thanks to everyone who contributed.

Read the full release notes: flower.ai/docs/framework…

@chongshenng is joining AI 2026 BEYOND today with a live Flower Endeavor demo and Q&A.

Join us online at 09:00 GMT / 18:00 KST to see Endeavor in action and bring your questions.

Event details: promotion.zep.us/ai-beyond-2026…

Flower Monthly November x.com/i/broadcasts/2…

Choose where it runs. Inspect its thinking traces. Keep its outputs and use them to train your own models.

That’s what “under your control” means with Endeavor 1.1.

Request access in the launch thread below.

x.com/flwrlabs/statu…