WorkBuddy Bench (Claude Code) - Web: leaderboard
Tencent WorkBuddy Bench (2026) evaluates coding agents on colloquial, role-played requests reverse-engineered from real work. This board is the Web track, 70 front-end and GUI tasks scored from 0 to 100 by rules, LLM or VLM judges and agent rubrics, with each model run in the Claude Code harness in think mode, mean of three runs; higher is better.
Metric: Score (0-100) on the 70 front-end and GUI tasks scored by rule, LLM or VLM and agent rubrics, run in the Claude Code harness in think mode, mean of three runs; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 7 models tracked. Updated 30 Sep 2026.
Top models
| # | Model | Setting | Score | Overall rank |
|---|---|---|---|---|
| 1 | Claude Opus 4.8 | Thinking | 69.86 | #58 |
| 2 | Hy3 | 66.43 | #199 | |
| 3 | GPT-5.5 | 64.86 | #52 | |
| 4 | GLM-5.2 | 60.71 | #141 | |
| 5 | MiniMax-M3 | 52.57 | #272 | |
| 6 | DeepSeek V4 Pro | Thinking | 51.57 | #191 |
| 7 | DeepSeek V4 Flash | Thinking | 50.29 | #259 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=workbuddy-bench-claude-code-web · How It Works · Data refreshed daily, snapshot 2026-11-16.