Apodex 1.1 mini targets long-horizon agentic work

Apodex has published Apodex 1.1 mini, a text and reasoning model built on the Qwen3.5-35B-A3B base and released on Hugging Face. The project frames the model as an agentic system, tuned for long-horizon, multi-step tasks rather than quick single-turn responses.

The model uses a mixture-of-experts (MoE) design, an architecture that activates only a fraction of its parameters per token. That approach keeps inference costs closer to a smaller dense model while retaining a larger pool of specialized capacity — a common choice for teams that want reasoning ability without paying full price on every forward pass.

Why it matters

The pitch here is agentic durability: models that can hold a plan together across many steps, tool calls, and intermediate decisions. Much of the recent open-weights work has emphasized exactly this shift.

• Built on the Qwen3.5-35B-A3B base, itself part of Alibaba's Qwen line
• MoE architecture for efficiency at inference time
• Positioned for reasoning and long-horizon task completion

Some details remain unspecified in the release record, including active parameter count and context length, and the model ships under an unspecified "other" license — worth checking before any commercial use. Supporting material is linked from the associated paper. As always with fast-moving open releases, independent evaluation will determine whether the agentic claims hold up in practice.