Qwen 3.6 27B Makes Huge Gains in Agency on Artificial Analysis - Ties with Sonnet 4.6

It is crazy that Qwen3.6 27B now matches Sonnet 4.6 on AA's Agentic Index, overtaking Gemini 3.1 Pro Preview, GPT 5.2 and 5.3 as well as MiniMax 2.7. It made gains across all three indices but the way the Coding Index works, I don't think the gains are as apparent as they should be. The Coding Index only uses Terminal Bench Hard and SciCode which are both strange choices. Cleary the training on the 3.6 models out now has focused on agentic use for OpenClaw/Hermes but it's interesting how close t