AI Wiki · Mediocre Security
AI labFrontier labs#research

IQuest

Model builder whose IQuest-Q1 (320B sparse MoE) claims top agentic-coding scores and 30 vulnerability reports from a three-day live deployment.

IQuest published its IQuest-Q1 report (arXiv 2610.30949) on Oct 22 2026: a 320B sparse MoE with 15B active parameters, released as an open-source model. AI Weekly, relaying the paper on Hugging Face, reports a 512K context and scores of 83.2 on Terminal-Bench 2.1, 57.5 on SWE-Bench Pro and 93.3 on GPQA Diamond, but much lower scores on newer Terminal-Bench 3.0 (17.9) and 4.0 (22.7). It also says human security experts assisted by the model produced 30 independently validated vulnerability reports during a three-day deployment, though the paper lists no CVEs or projects. All figures are self-reported and we did not read the full paper; the abstract page does not describe who IQuest is. Treat as unverified until reproduced.

Connections

IQuest and everything one link away. Drag to move, scroll to zoom, click a node to open it.Open in the full graph →

What links here

Concept Open-weight vs. closed models

Sources

  1. arXiv, “IQuest-Q1: Advancing Foundation Capabilities for Agentic CLI Systems”, Oct 22 2026
  2. AI Weekly, “IQuest-Q1 opens 320B/15B MoE, 83.2 on Terminal-Bench 2.1”, Oct 23 2026 (citing the Hugging Face paper page)