AIMultiple - TweetEval Offensive: leaderboard
AIMultiple zero-shot TweetEval Offensive: classify the first 200 test tweets as offensive or not with a single-label reply at temperature 0. Scored with macro-F1 as defined by TweetEval and published as whole numbers; %, higher is better.
Metric: Macro-F1 (%). Source: aimultiple.com. Saturation forecast: Around April 2027. 10 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.5 | 73 | #33 |
| 2 | Claude Sonnet 4.6 | 72 | #94 |
| 3 | Claude Opus 4.8 | 70 | #39 |
| 4 | Gemini 3.1 Pro (Preview) | 69 | #64 |
| 5 | Kimi K2.6 | 69 | #113 |
| 6 | Qwen 3.6 Plus | 69 | #124 |
| 7 | GLM-5.1 | 69 | #109 |
| 8 | Gemini 3.5 Flash | 68 | #66 |
| 9 | MiniMax-M2.7 | 65 | #270 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=aimultiple-tweeteval-offensive · How It Works · Data refreshed daily, snapshot 2026-11-05.