YecoAI

Ender-1 Enters Beta: Our 1.5B Model Trained on EnderDevelopment Data

Ender-1 is a fine-tune of Qwen2.5-Coder-1.5B (Apache 2.0) trained on EnderDevelopment data. In beta, with a staged rollout in shadow mode.

Marco Nasi

Founder & CEO

Correction: an earlier version of this article described Ender-1 as based on Qwen2.5-Coder-3B. That was an evaluation error: the 3B model has a non-commercial license. Ender-1 is based on Qwen2.5-Coder-1.5B, released under Apache 2.0.

Today Ender-1 enters beta. It is our first model born from a live product: it is trained on EnderDevelopment data, the gaming platform we already scaled to over 30,000 registered users (figure updated continuously, so it may not always be current when you read it).

What Ender-1 does

Ender-1 is a fine-tune of Qwen2.5-Coder-1.5B, purpose-built for the gaming ecosystem. It generates Minecraft (Spigot/Paper) plugins, FiveM scripts, and Discord bots from natural-language intent, and repairs server configurations (YAML/JSON, BungeeCord/Velocity) when they break.

Trained on our data, within policy

The training corpus comes from EnderDevelopment free-tier activity, strictly within platform policy: feature requests, accepted outputs, compile results, and community-vetted configurations. No private or paid-tier data is used, ever.

Rolling out now

Ender-1 is in beta and beginning its staged rollout on EnderDevelopment in shadow mode. Real-world acceptance signals from the rollout feed the next training cycle, the same compact-model, rigorous-evaluation methodology behind Yeco-PII.

The gaming platform we scaled is now the training ground for our own model. That is our thesis: every product generates the data that makes the next model better.

Related content