Flower Labs

Releases Endeavor 1.1, adding image inputs and the option to run the model on GPUs in the EU; it headlines a composite score narrowly ahead of OpenAI's GPT-5.6 Sol, but the figures are self-reported and built on tests where top models already score near the maximum.

Opens early-access preview of Flower SuperGrid, a platform for setting up and managing federated-learning networks at scale and a step toward a paid product built on the free framework.

Launches Endeavor 1.0, a general-purpose model customers can run on their own servers, as a preview for selected organizations; Flower says it rivals OpenAI and Anthropic on some tests, while press reports say it draws on open models from Meta and Mistral and that the company expects to raise money again in about a year.

With data-network startup Vana, trains Collective-1, a 7-billion-parameter language model, on GPUs spread across the internet rather than in one data center; co-founder Nic Lane tells Wired the company is training a 30-billion-parameter model and plans a 100-billion-parameter one.

Releases a preview of Lizzy 7B, a small open-weight model (its files are free to download) built in the UK for public-sector and enterprise use, with company-run benchmarks showing it ahead of comparable European models on most tests.

Launches Flower Intelligence in preview, a service that runs AI on users' own devices and hands off to Flower's encrypted cloud when more computing power is needed; Mozilla signs on to use it for an AI summary feature in its Thunderbird email app.

Launches as a company, with Y Combinator as an early backer, to commercialize the Flower open-source project; in August it announces a $3.6 million pre-seed round.

Red Hat and Flower Labs engineers publish an open-source add-on that lets companies deploy and manage Flower across many groups of servers through Open Cluster Management, the open-source project underlying Red Hat's cluster-management product.

Raises a $20 million Series A led by Felicis at a reported $100 million valuation, with 13 employees, saying it is not yet focused on revenue and will build paid tools on top of its free framework.