Endeavor 1.0

// Overview

Endeavor 1.0 is Flower Labs' first production-ready Endeavor model, introduced on September 1, 2026, four months after Flower's 7B Lizzy model for UK use. Flower calls it a frontier-class generalist with strong reasoning and coding, designed for the long-horizon work carried out by agents. It launched as a preview for a select group of organizations and partners rather than as open self-service access.

Flower's launch evaluation gives it 92.0 on GPQA, 98.2 on HumanEval, 99.9 on AIME 2026 and 94.1 on IFEval. In the same comparison GPT-5.6 Sol scores 94.1, 95.1, 99.9 and 95.9, and Claude Fable 5 92.6, 97.0, 99.9 and 91.7: Endeavor 1.0 has the highest HumanEval score, matches both on AIME 2026, beats Kimi K3 on three of the four benchmarks and Nemotron 3 Ultra on all four. These are vendor-reported figures.

Flower says Endeavor builds on capabilities from the open-weight ecosystem plus its own work, and that much of the effort went into how the model works with its harness: how it allocates reasoning effort, builds and keeps context, uses tools and recovers when a step fails. Its evaluation design also draws on FlowerBench, Flower's benchmark of long-horizon enterprise tasks run inside participating organizations' own environments. The model is offered as a production service operated by Flower or deployed inside an organization's own infrastructure, and a team can mix the two.

Endeavor 1.1 followed on November 3, 2026, adding image input and EU-located inference.

// Benchmarks

This model's scores

1. GPQA92%
2. HumanEval98.2%
3. AIME 202699.9%
4. IFEval94.1%

Scores on a 0–100 scale (25-point gridlines); higher is better. Each benchmark links to its published source.

// Strengths

• 98.2 on HumanEval, the top score in Flower's launch comparison
• 99.9 on AIME 2026, level with GPT-5.6 Sol and Claude Fable 5 in that comparison
• 92.0 on GPQA and 94.1 on IFEval
• Runs as a Flower-managed service or inside an organization's own infrastructure
• Harness work on reasoning effort, context handling, tool use and error recovery for long-running agents

// Best for

• Reach for Endeavor 1.0 as the core generalist behind agents and complex workflows that must be deployable in your own environment.
• Reach for it when some sensitive applications must run privately while the rest use a managed service.
• Look elsewhere when you need image input: that arrived with Endeavor 1.1.

// Endeavor — every version

The full lineage of the Endeavor line, newest first. Every version has its own page — click any to compare specs, benchmarks and pricing.