Reddit hands-on: Qwen3.5 9B struggles at agentic work while 35B A3B handles it
Nautisop · reddit · 2026-10-27
A Reddit user compared Qwen3.5 sizes on agentic tasks: in goose with the Microsoft Learn MCP and a matching skill, the 35B A3B model requests data flawlessly (if slowly), while 9B fails to recognize the skill unless explicitly told to "use docs," then misuses the API despite it being described in the skill. The poster questions claims that 9B can handle agentic work — in practice it was very weak.
More from coding & agent
- Compound Engineering 3.33.0 ships PR workflow upgrades to curb scope-creeping agents — kieranklaassen · 2026-10-27
- Lovable invokes 753.6B Cloudflare Workers monthly, migrated from Next.js to TanStack Start — ritakozlov · 2026-10-27
- Harness Lens now indexes 30 open source coding harnesses and accepts new submissions — yb2698 · 2026-10-27
- Prompting agents at the grammar level: pattern language as interface design — DanGrover · 2026-10-27
- Microsoft Foundry hosted agents get mailboxes, identity and org-chart presence — marlene_zw · 2026-10-27
- 36 blind AI judges pick a solo-session agent setup over Superpowers 36-0 — Kai_ThoughtArchitect · 2026-10-27