A research team published a technical report on Apodex 1.1, a language model trained to carry out long-running work involving files, web search and running code rather than single questions. The authors say it reaches the leading performance band on finance, science, maths, coding and search tasks while using a smaller model than many rivals. A 35-billion-parameter Mini version is described as deployable on local hardware; the report states results without independent verification.
What changed
Earlier versions of the model were trained on narrower task environments and less coordination between agents.
What it unlocks
Running a model that sustains multi-step work with files, search and code in a locally deployable 35B size.
- 35B parameters for Mini version
Sources