Flower Labs emblem on a white background, for the Endeavor 1.1 guide

Endeavor 1.1: Power-User Guide to Flower Labs’ Frontier Model, API and Setup

by

Last updated: November 3, 2026

Endeavor 1.1 is Flower Labs’ new frontier model, released November 3, 2026. Compared with Endeavor 1.0 it adds image input, an EU-only inference option and stronger reasoning and coding. It also keeps readable thinking traces and output ownership. Access is by request through an OpenAI-style Responses API or private deployment.

Contents

Flower Labs is best known for its open-source federated learning framework. Endeavor 1.1 is its third model release this year, after Lizzy and Endeavor 1.0, according to the Endeavor 1.1 launch post. The company positions it as a model you can call as a managed service or run on infrastructure you control.

What’s new in Endeavor 1.1

The launch post lists these changes and features:

Feature What Flower says
EU inference New option to run inference on GPUs located in the European Union, alongside the global offering
Multimodal input Image inputs are now supported alongside text
Reasoning and coding Stronger than Endeavor 1.0; Flower reports GPQA rising from 92.0 to 94.4 and HumanEval from 98.2 to 99.9
Thinking traces Readable traces you can inspect to review reasoning and investigate decisions
Output ownership You keep outputs and traces and may use them to train your own models
Deployment Flower-managed service, private cloud or on premises
Support Service-level agreements and deployment support from Flower

Flower says the GPQA gain works out to about 30% fewer errors, and that it sees similar gains in cybersecurity analysis and complex agent workflows. These benchmark figures are Flower’s own; no independent evaluation was published with the launch.

Endeavor 1.1 vs Endeavor 1.0

Both versions remain available by request, according to the Endeavor model documentation. The client setup guides use 1.1 in their examples.

Aspect Endeavor 1.1 Endeavor 1.0
Status Latest release Previous release
Input types Text and images Text
Inference regions Global or EU GPUs Global
GPQA (Flower-reported) 94.4 92.0
HumanEval (Flower-reported) 99.9 98.2
Model name flwrlabs/endeavor-1.1 Confirmed during onboarding

The “Text” entry for 1.0 reflects Flower’s statement that image input is new in 1.1. The EU entry for 1.0 reflects Flower’s statement that EU inference starts with 1.1.

Access, model ID and context limits

The documentation gives these details for the managed service:

  • Base URL: https://api.flower.ai/v1
  • Model name: flwrlabs/endeavor-1.1
  • API style: the Codex and OpenCode guides use the Responses wire format, so OpenAI-compatible clients that speak Responses can connect.
  • Credentials: a Flower API key, read from the FLOWER_API_KEY environment variable in the client guides.
  • Context: up to 1M tokens supported; the supplied Codex catalog uses 128,000 tokens with compaction at 96,000. The effective limit depends on your deployment.

Flower does not publish per-token pricing. Access goes through a request form, and onboarding confirms the endpoint, supported response formats, tool and streaming support, usage limits and support channels. If you are budgeting against published price lists, our Mistral API pricing breakdown covers another European option with public rates.

Flower also says it plans UK-only inference and regional options in the Middle East, Asia and Africa. For other European model efforts, see our note on Aleph Alpha’s Kolibri model for public infrastructure.

How to use Endeavor 1.1 with Codex

Flower’s Codex setup guide covers the Codex CLI and local Codex tasks in the ChatGPT desktop app on macOS:

  1. Download Flower’s flower-models.json model catalog and copy it to ~/.codex/flower-models.json.
  2. Back up ~/.codex/config.toml, then set model = "flwrlabs/endeavor-1.1", model_provider = "flower", model_reasoning_effort = "low" and model_catalog_json at the top level.
  3. Add a [model_providers.flower] section with base_url = "https://api.flower.ai/v1", wire_api = "responses" and env_key = "FLOWER_API_KEY".
  4. Export FLOWER_API_KEY in your terminal.
  5. Run codex in your project directory. Endeavor is selected by default.
  6. For the desktop app, quit it fully and relaunch from the same terminal with open -a ChatGPT --env FLOWER_API_KEY="$FLOWER_API_KEY", then pick Endeavor in the model selector for a local task.

Flower warns that opening the desktop app from the Dock may not pass the key, so relaunch from the terminal each time.

How to use Endeavor 1.1 with OpenCode

The OpenCode setup guide adds a flower-labs provider to ~/.config/opencode/opencode.json using the @ai-sdk/openai package, which Flower says is required because it uses the Responses API. Start it with opencode --model flower-labs/flwrlabs/endeavor-1.1. Our OpenCode 2.2 guide covers the client’s newest features.

Workflows that fit Endeavor 1.1

  • Trace review for agent failures. When an agent run goes wrong, read the thinking trace to find the step where the plan broke, instead of guessing from the final output.
  • Distilling a specialist model. Flower’s own example is an engineering team that collects verified fixes and their traces, then trains a smaller troubleshooting model on them. Keep only reviewed examples.
  • EU data residency. Request the EU inference location when workloads must stay on EU GPUs.
  • Private deployment for sensitive code. Run the model in your own cloud or on premises, and choose when to adopt upgrades.
  • Screenshot and diagram input. Use the new image input for UI bugs, charts and architecture diagrams.

For self-hosted serving stacks more broadly, see our vLLM 0.32 guide. For another November release aimed at long agent tasks, see our LongCat-2.5 guide.

Known limitations

  • There is no self-service sign-up or public price list; you must request access.
  • The setup guides cover macOS and zsh only.
  • Model artifacts for private deployment are distributed privately, and formats are confirmed during onboarding.
  • Benchmark results are reported by Flower Labs and compare against its own selected composite of GPQA, AIME 2026, HumanEval and IFEval.

Frequently asked questions

What is Endeavor 1.1?

Endeavor 1.1 is a frontier-class language model from Flower Labs, released on November 3, 2026. It is built for reasoning, coding and long-horizon agent work, and it adds image input, an EU inference option and readable thinking traces compared with Endeavor 1.0.

How do I get access to Endeavor 1.1?

Access is by request. Flower asks organizations to fill out the Endeavor access form, then issues a Flower API key during onboarding. The managed service uses the base URL https://api.flower.ai/v1 and the model name flwrlabs/endeavor-1.1.

How much does Endeavor 1.1 cost?

Flower Labs does not publish per-token prices on its Endeavor documentation or model page. The model is offered by request, so pricing and usage limits are set during onboarding for the managed service or a private deployment.

What is the Endeavor 1.1 context window?

Flower says Endeavor supports up to a 1M-token context window. Its supplied Codex configuration uses 128,000 tokens with automatic compaction at 96,000 tokens, and the effective API limit depends on your deployment.

Can I run Endeavor 1.1 on my own servers?

Yes. Flower offers private deployment in your own cloud or on premises, with Flower support. Model artifacts are distributed privately, and formats are confirmed during onboarding.

Can I use Endeavor 1.1 outputs to train my own models?

Yes. Flower says customers own their inference outputs and thinking traces and may use them to fine-tune or train their own models, including smaller specialist models.

Sources

Image: Flower emblem from the Flower open-source repository, Apache License 2.0, via GitHub. Flower and Endeavor are trademarks of Flower Labs. FreeAnswers is not affiliated with or endorsed by Flower Labs.



Related answers

Leave a Comment