AionLabs: Aion 3.5 Mini

Aion 3.5 Mini is a hosted text model from Aion Labs, released on September 23, 2026, as the lower-cost companion to Aion 3.5. It is described as a multi-model roleplaying and storytelling system built on the GLM family: several specialized models collaborate on each reply to improve narrative structure, tension, conflict, and nuanced handling of mature themes. The service is accessed only through APIs such as Aion Labs and OpenRouter (model id aion-labs/aion-3.5-mini); weights are not published and there is no Hugging Face model card with standard academic benchmarks.

Operationally, Aion 3.5 Mini offers a 262,144-token context window, up to 32,768 completion tokens, mandatory reasoning with configurable effort (low, high, or max; default high), tool calling, structured JSON via response_format, and prompt caching at reduced read pricing. List pricing is about $0.70 per million input tokens and $1.40 per million output tokens. OpenRouter telemetry reports on the order of ~59 tokens per second median throughput for the hosted endpoint.

As of public documentation and major benchmark aggregators, Aion Labs does not publish an official evaluation table for Aion 3.5 Mini, and independent trackers (for example Artificial Analysis and composite quality indexes) list no verified MMLU, GPQA, coding, or agentic scores for this exact model. It is widely used in production for interactive fiction and character chat (e.g., Janitor AI and SillyTavern traffic on OpenRouter), but task benchmark numbers should be treated as unavailable until the vendor or a named third-party evaluation reports them.