# IQuest Research releases open-source Q1 model and 30‑benchmark report

> The new IQuest‑Q1 foundation model, built from scratch, uses a 320 B Mixture‑of‑Experts design and reports results on 30 coding and agentic benchmarks.

Oossa · 2026-10-23 · https://oossa.com/en/iquest-research-releases-open-source-q1-model-and-30-benchmark-report

IQuest Research published the IQuest‑Q1 technical report on 23 Oct 2026. The model is open‑source, optimized for command‑line tasks, and has 320 billion parameters total, with 15 billion active per token. Training covered about 29 trillion tokens and claimed 52 % of the compute needed by a comparable MoE baseline. The report lists scores on 30 benchmarks, including 52.7 on AutomationBench and 78.4 on MCP‑Atlas.

## The facts

- 320 billion total parameters, 15 billion active per token
- Training used ~29 trillion tokens and achieved 1.93× scaling efficiency

## Why it matters

Developers can now download a large, agentic CLI model that promises lower compute costs for comparable performance.

## Sources & references

1. [IQuest Research Releases IQuest-Q1 Technical Report, Detailing Training from Scratch and Results Across 30 Benchmarks](https://www.prnewswire.com/news-releases/iquest-research-releases-iquest-q1-technical-report-detailing-training-from-scratch-and-results-across-30-benchmarks-302917967.html) – PR Newswire RSS, 2026-10-23

Last updated: 2026-10-23