🥇 Claude 3.5 Sonnet
46 / 50 · 1.0‰ cent per runHook strength
9/10
Specificity
9/10
Char compliance
10/10
Natural voice
9/10
Would plausibly perform
9/10
Opened with "I replaced 3 Chrome extensions this quarter." Specific, grounded, high-signal.
Task: Make a boring tweet actually interesting.
Claude 3.5 Sonnet wins on text craft, but Grok is a legitimately good X/Twitter voice generator — it was trained on tweets, and it shows. For a brand voice at scale, Grok's "native register" is a real differentiator worth testing.
Rewrite this tweet to be more engaging while staying under 280 chars and preserving the main claim. Original: "AI tools can help you save time at work. Here are 5 tools that have changed how I work this year." Rules: - Keep it under 280 chars - One concrete hook in the first 10 words - At least one specific detail (number, name, timeframe) - Written to perform — but not clickbait that misleads - No emoji unless it adds signal - Written like a person tweets, not like a brand
| Rank | Model | Total /50 | Cost / run |
|---|---|---|---|
| 🥇 | Claude 3.5 Sonnet | 46 | 1.0‰ cent |
| 🥈 | Grok | 44 | 3.0‰ cent |
| 🥉 | GPT-4o | 38 | 2.0‰ cent |
| #4 | Gemini 2.0 Flash | 37 | 0.3‰ cent |
Opened with "I replaced 3 Chrome extensions this quarter." Specific, grounded, high-signal.
Native Twitter/X voice. Dropped the "5" and anchored around 3 — tighter.
Got the structure right but ended with "Which one should I share first?" — a mild engagement-bait tell.
Fine but generic. Over-used "transformed" (close to a banned AI-ism).