V2EX / 2026-10-28

[分享创造] 做了个让 DeepSeek 操作浏览器和手机的 harness： V4.1 Flash 跑 MobileWorld 96/117、WebArena-Verified 98/111

让 DeepSeek 操作浏览器、Android 和 iPhone ，用 V4.1 Flash 。核心是给每一步加检查：规划 → 执行 → 前后截图对比验证。点了没生效、输入的内容不对，当步就能发现并重新规划，所以便宜的 flash 模型也能把事情做对。比如让它通关一局五子棋：自己选人机对战、改难度、落子，直到获胜。基准成绩（全程只用 V4.1 Flash ，按官方评测器判分）：MobileWorld 96/117 ，WebArena-Verified 98/111 。仓库里有每个基准 10 条抽样运行的回放和评分。源码整理中，整理完开源（ MIT ）。 https://github.com/hyechow/seek-use

Read original source

Source page screenshot: [分享创造] 做了个让 DeepSeek 操作浏览器和手机的 harness： V4.1 Flash 跑 MobileWorld 96/117、WebArena-Verified 98/111 — Source page screenshot · www.v2ex.com2026-11-06 UTC

Source record

Source: V2EX
Published: 2026-10-28 02:24:09 UTC
First discovered: 2026-10-28 02:45:18.355Z
Last collected: 2026-10-28 02:45:18.355Z
Author: glayyy8717
Topic: Developer tools
Classification: Open-source projects & tools
Evidence level: E1 · source record; conclusions unverified

Research analysis

Rule-based research guidance derived from the source record. Audience and commercial demand require validation.

Project record

Source account: 让 DeepSeek 操作浏览器、Android 和 iPhone ，用 V4.1 Flash 。核心是给每一步加检查：规划 → 执行 → 前后截图对比验证。点了没生效、输入的内容不对，当步就能发现并重新规划，所以便宜的 flash 模型也能把事情做对。比如让它通关一局五子棋：自己选人机对战、改难度、落子，直到获胜。基准成绩（全程只用 V4.1 Flash ，按官方评测器判分）：MobileWorld 96/117 ，WebArena-Verified 98/111 。仓库里有每个基准 10 条抽样运行的回放和评分。源码整理中，整理完开源（ MIT ）。 https://github.com/hyechow/seek-use

Observed adoption & comparisons

Recent records to compare: “astral-sh/ruff” (GitHub Trending, Nov 6, 2026); “microsoft/vscode” (GitHub Trending, Nov 6, 2026). Shared themes are navigation hints; they have not been verified as the same need or independent corroboration.

Adoption decision

Evaluate it against the tool or manual workflow it would replace. The stored record supports discovery; it does not yet establish reliability, lower operating cost or commercial adoption.

Unverified requirements

Not verified: license permissions, required hardware, installation steps, maintenance status and a representative task with its failure cases.

Next test

Choose one real task and a current alternative. Record the commit, environment, input, output, elapsed time and failure behavior; retain the result before recommending adoption.