AIMultiple - TweetEval Average: leaderboard
AIMultiple's zero-shot TweetEval run: the first 200 test tweets of five tasks (emotion, hate, irony, offensive, sentiment), temperature 0 through OpenRouter, each scored with the TweetEval authors' official metric. This series is the mean of the five task scores, published as whole numbers; %, higher is better.
Metric: Mean of the five TweetEval task scores (%). Source: aimultiple.com. Saturation forecast: Around February 2027. 10 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.5 | 80 | #33 |
| 2 | Claude Sonnet 4.6 | 79 | #94 |
| 3 | Qwen 3.6 Plus | 78 | #124 |
| 4 | Kimi K2.6 | 77 | #113 |
| 5 | Gemini 3.1 Pro (Preview) | 76 | #64 |
| 6 | GLM-5.1 | 76 | #109 |
| 7 | Claude Opus 4.8 | 74 | #39 |
| 8 | Gemini 3.5 Flash | 73 | #66 |
| 9 | MiniMax-M2.7 | 72 | #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-average · How It Works · Data refreshed daily, snapshot 2026-11-05.