AllSpark releases Iris-mini and Iris-pro, top open-weight search agents in their class
The Decoder · rss · 2026-09-13
The AllSpark team has released Iris-mini and Iris-pro, two open-source search agents built on Qwen models that top benchmarks among open-weight models in their size classes. Per the paper, the training data and models also improve performance on tasks they were never trained for, including general tool use and office work.
More from coding & agent
- Mac MCP: open-source MCP server turns ordinary ChatGPT chats into a macOS control plane — bulutarkan · 2026-09-13
- Developer hands his trivia site to a Grok bot that ships new content and SEO fixes daily — santiviquez · 2026-09-13
- Manus agent builds a sourced 10-tool comparison deck from a single prompt in under 20 minutes — Aiden_Tech_Ai · 2026-09-13
- Dev builds px0, a read-only IDE that boots in 1ms and idles at 16MB vs VS Code's 1.4GB — arpit_bhayani · 2026-09-13
- OpenAI hosts GPT-6 Astra hackathon; attendees say GPT-Live-1 blew their minds — gabrielchua · 2026-09-13
- Ex-Microsoft exec: AI coding agents cut my setup from 3 monitors to 1 laptop — MParakhin · 2026-09-13