Catalog
We start with the best open models and fine-tune them for the ecosystem: CodiiX, agents, APIs, and personal nodes. The same quality ambition as other providers — with the power of running local.
Available now
From the smallest for the whole company to Frontier — open downloads from Hugging Face, and OpenRouter when it is worth it.
Up to 290 t/s
Built on Gemma 4 E4B. Casual conversations, document reading, and small agents. Share real AI across the company at low cost. Handles email and connects with XMail (mail + calendar). Up to 290 t/s depending on GPU.
Up to 70 t/s
Built on Qwen3.8-27B and fine-tuned for CodiiX, Work, Programming, Agents, and Architecture. Runs on personal nodes with long context, vision, and tool calling.
Apache 2.0 (Qwen). LICENSE / NOTICE.
Up to 200 t/s
Built on Qwen3.6-35B-A3B-MTP. Expert in architecture and DevOps, manages all kinds of agents and document vision. Our fastest fine-tune: up to 200 t/s depending on GPU.
Apache 2.0 (Qwen). LICENSE / NOTICE.
Up to 40 t/s
Built on Qwen3.8-Flash-Next. Frontier intelligence vs the 27B: infernal bugs, 3 reasoning levels, canvas and 3D, MCPs, polished finishes and landing pages. Up to 40 t/s on 2× RTX 3090.
Qwen Community License 1.0: Titan is shared for whatever use you want on Xeretron. View disclaimer.
Up to 18 t/s
Built on GLM-5.3-Flash (Z.ai). Competes with Frontier models: top-tier architectures, impossible bugs, device control, and agent orchestration. Up to 18 t/s on 2× RTX 3090 · MIT.
MIT (Z.AI). LICENSE / NOTICE.
Xeretron lets you download models from Hugging Face and experiment openly from the panel to use them however you want — beyond the official catalog.
We are integrated with OpenRouter to use external power when you need it and giant models when they are truly required and worth it. Local remains the default.
How we choose
We start from open families with real results in coding, work, and agents — not demos.
Each model is trained or aligned for CodiiX, the panel, APIs, and the flows you already use on Xeretron.
We size VRAM, context, and quantization with you. If it will not fit well, we tell you upfront.
Catalog
Junior for the whole company and XMail. Flagships when the work demands it. HF to experiment. OpenRouter when it is truly worth it.
