Local Agent trend: JetBrains and Perplexity optimize for on-device inference

ycombinator · x · 2026-08-28

Local deployment trends emerged this week with JetBrains and Perplexity launching local agents based on the 27B Qwen model. The key challenge lies in optimizing agent behavior on consumer hardware. JetBrains found that agents spend most of their time reading, leading Junie Local to adopt the M5 chip for a 40% speed boost in reading tasks. Additionally, tests indicated that the Qwen 3.8B model requires 'thinking' mode enabled for reliable behavior.

Original post →

More from coding & agent

coding & agent channel →