# AllSpark releases Iris open-weight web-search agents

_Sunday, September 13, 2026 at 8:58 AM EDT · AI · Latest · Tier 2 — Notable_

![AllSpark releases Iris open-weight web-search agents — Primary](https://the-decoder.com/wp-content/uploads/2026/09/search-agents-allspark-nano-banana-pro.jpg)

Chinese lab AllSpark released Iris-mini and Iris-pro, open-weight web-search agents based on Qwen-series models, with a 256,000-token context window.

The team also released an Iris Harness containing the agent loop, tools, context-management strategies and evaluations; it says the harness can run against any OpenAI-compatible endpoint. According to the paper, Iris-mini has 35 billion parameters and Iris-pro has 397 billion, and both lead their open-weight size classes on several reported benchmarks.

The training-data construction and training pipelines are planned for a later release.

## Sources

- [the-decoder.com](https://the-decoder.com/iris-mini-and-iris-pro-are-the-strongest-open-weight-search-agents-in-their-class/)

---
Canonical: https://techandbusiness.org/newswire/py8iVz49j4xVXYVT1LUyxW
Retrieved: 2026-09-13T17:15:21.237Z
Publisher: Tech & Business (techandbusiness.org)