IntentGrasp - Toxic Speech: leaderboard

The Toxic Speech domain of the IntentGrasp (2026) All Set, intent-understanding items drawn from public intent datasets and posed as multiple choice with one or more correct intents. The score is instance F1 in percent, averaged over option-order shuffles; higher is better.

Metric: Instance F1 (%) on the Toxic Speech (TS) domain of the 12,909-item All Set, mean over option-order shuffles, zero-shot at temperature 0; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 11 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Gemma 4 31B (IT) [4bit]55.41#1803
2Qwen3-32B [4bit]50.5#1273
3gemma-4-E4B-it49.94#1539
4Olmo-3.1-32B-Instruct [4bit]48.1
5Gemma 4 E2B (IT)47.35#2451
6Llama 3.3 70B Instruct [4bit]44.11#1257
7Qwen 3 4B42.31#2348
8Qwen 3 8B38.97#1873
9OLMo 3 7B Instruct37.24#6953
10Llama 3.1 8B Instruct29.29#4600
11Llama 3.2 3B Instruct24.48#7241

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=intentgrasp-toxic-speech · How It Works · Data refreshed daily, snapshot 2026-11-21.