Aion 3.5

• 🧠 Multi-model system: several specialized models collaborate on each response
• 📚 Built for roleplaying, storytelling, narrative structure, tension and conflict
• 📏 262,144-token context window as listed in this catalog
• 🆕 Released September 2026, alongside a Mini sibling
• 🔧 Supports function calling for interactive fiction and tool-driven scenes
• 🎯 Listed with reasoning among its capabilities
• 🌐 Web search available through this catalog's integration
• 🔒 No published parameter count, license, or quantization details

Aion Labs

Aion Labs is an Israeli venture studio working at the intersection of artificial intelligence and pharmaceutical research. The group focuses on applying machine learning to drug discovery and development, partnering with scientific and industry collaborators…

Aion 3.5 is the flagship text model in Aion Labs' Aion line, aimed at roleplaying and long-form interactive fiction rather than general assistant work. It is described as a multi-model system built on the GLM family: instead of a single network answering, multiple specialized models collaborate on each response, an arrangement intended to produce stronger narrative structure and more compelling tension and conflict. It is also described as handling mature and darker themes with nuance.

It succeeds Aion 3.0, which was released in July 2026, and arrives in September 2026 together with Aion 3.5 Mini, the smaller tier of the same generation. The earlier generation likewise paired a flagship with Aion 3.0 Mini, so the lineup keeps a two-tier shape across both releases.

In this catalog, Aion 3.5 is listed with a 262,144-token context window and three capabilities: reasoning, function calling, and web search. The long window suits extended chat sessions and multi-chapter stories where earlier turns need to stay in view, while function calling allows external tools to be wired into a scene. No parameter count, license, or quantization information is published for the system, which is consistent with it being a proprietary ensemble rather than an open weights release. Aion Labs does not publish a benchmark table for this generation, so its narrative-quality positioning should be read as a vendor description.