Local execution of Qwen3.8-Flash-Next paired with Claude Code for code review

On 2026-11-03, a user in the Reddit LocalLLaMA community shared a hands-on experience running Qwen3.8-Flash-Next locally while using Claude Code as the driving agent for code review. The test target was a real commit in the repo at https://github.com/leok7v/ChatOKF, a Swift on-device LLM chat app. That commit touched roughly 48 files, adding 1048 lines and deleting 102, covering a module move plus a new 783-line search harness. The setup ran on AMD Strix Halo hardware with 128 GB of memory, loading the Qwen3.8-Flash-Next UD-Q4_K_XL quantized build from unsloth through llama-server, with ROCm providing local inference, a 262K context window, an MTP draft head enabled, and Claude Code driving the whole pipeline in headless mode. The author reported being seriously impressed by the result, and ran the same review twice on the same commit to compare.

A second post appeared in the same time window on V2EX, which is unrelated to the Qwen3.8-Flash-Next testing. In that thread, a developer described shipping a browser extension built with AI assistance that automatically saves text drafts on sites where protection is enabled. The extension stores drafts locally, requires no account registration, uploads nothing, and skips password fields plus other sensitive entries such as verification codes, payment data, and ID documents. It is meant for scenarios like filling out web forms, editing product listings, or writing comments, where content could otherwise be lost to accidental deletion, page refresh, or a closed editor window. The developer noted that Google's extension review was quick: submitted one day, approved the next. The listing is published on the Chrome Web Store at https://chromewebstore.google.com/detail/jhfcpciejngnhokpodcggoagcoangekn. As of the reports, both items rest at individual user or developer shares, with no independent benchmark or broader follow-up evidence in the material.