Benchmark Almanac

AfriHate

AfroXLMR-76L 78.2% tops its leaderboard Gemma 2 27B 58.8% leads the models we track
Not in model cards

Models classify hate speech and abusive language across African languages.

AfroBench lists F1 as the metric for hate speech detection.

Released
January 2025
Kind
Reading in many languages
Level
Everyday
Scored
Checked automatically
Answers
Multiple choice
Tasks
All public
Measures
Capability
Risk
Hate
Language
Many languages
In and out
Text → Text
Interaction
Single turn
Refresh
One fixed set
Run by
Anyone
Strongest model tested
Llama 3.1-70B #177 on the capability index
Last model added
July 2024
Scores refreshed
November 20th, 2026
15
Entries in its paper, 5 models we track

AfroXLMR-76L tops the leaderboard with 78.2% Gemma 2 27B leads with 58.8% among models released by July 2024

Best score per entry from the benchmark's paper, read on November 20th, 2026. A paper’s table doesn’t add newer models. Where a model ran with several settings or harnesses, we keep its best. Every run is in the next chart. Names in gray aren't in our model list: fine-tunes, agent systems, products and older models. They don't count toward leaders. The best model we track is Gemma 2 27B, at 58.8%. This view shows only models we track, each at its best.

The newest model with a score here came out in July 2024, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:

  1. 1 AfroXLMR-76L Multilingual fine-tuning 78.2%
  2. 2 AfroXLMR Monolingual fine-tuning 76.2%
  3. 3 AfriBERTa Monolingual fine-tuning 72.3%
  4. 4 GPT-4o (snapshot unspecified) 20-shot 71.7%
  5. 5 AfriTeVa V2 Monolingual fine-tuning 68.7%
  6. 6 Gemma 2 27B 20-shot 58.8%
  7. 7 Llama 3.1-70B 20-shot 57.1%
  8. 8 Gemma 2 9B 20-shot 56.3%
  9. 9 LaBSE 20-shot 53%
  10. 10 aya-23-35B 20-shot 48.8%
  11. 11 Llama 3.1-8B 20-shot 42.7%
  12. 12 Mistral 7B v0.1 20-shot 41.4%
  13. 13 InkubaLM-0.4B 10-shot 25.4%
  14. 14 bloomz-7b1-mt 20-shot 25.4%
  15. 15 mt0-small 5-shot 22%

Source: The AfriHate paper

#EntryScoreSettingDateSource
1 AfroXLMR-76L 78.2% Multilingual fine-tuning Official leaderboard
2 AfroXLMR 76.2% Monolingual fine-tuning Official leaderboard
3 AfriBERTa 72.3% Monolingual fine-tuning Official leaderboard
4 GPT-4o (snapshot unspecified) 71.7% 20-shot Official leaderboard
5 AfriTeVa V2 68.7% Monolingual fine-tuning Official leaderboard
6 Gemma 2 27B 58.8% 20-shot Jun 2024 Official leaderboard
7 Llama 3.1-70B 57.1% 20-shot Jul 2024 Official leaderboard
8 Gemma 2 9B 56.3% 20-shot Jun 2024 Official leaderboard
9 LaBSE 53% 20-shot, SetFit Official leaderboard
10 aya-23-35B 48.8% 20-shot Official leaderboard
11 Llama 3.1-8B 42.7% 20-shot Jul 2024 Official leaderboard
12 Mistral 7B v0.1 41.4% 20-shot Sep 2023 Official leaderboard
13 InkubaLM-0.4B 25.4% 10-shot Official leaderboard
14 bloomz-7b1-mt 25.4% 20-shot Official leaderboard
15 mt0-small 22% 5-shot Official leaderboard
#ModelScoreSettingDateSource
1 Gemma 2 27B 58.8% 20-shot Jun 2024 Official leaderboard
2 Llama 3.1-70B 57.1% 20-shot Jul 2024 Official leaderboard
3 Gemma 2 9B 56.3% 20-shot Jun 2024 Official leaderboard
4 Llama 3.1-8B 42.7% 20-shot Jul 2024 Official leaderboard
5 Mistral 7B v0.1 41.4% 20-shot Sep 2023 Official leaderboard

Rankings by part

We read 15 parts from the benchmark’s paper, each ranking models on one part of the test, such as amh, ary and arq. Gemma 2 27B leads 10 of them, and Llama 3.1-70B leads the other 5. Arrows show how far a model moved from its overall rank among the models we track.

amh

5 models we track have a score here.

  1. 1 Gemma 2 27B 62.3%
  2. 2 Llama 3.1-70B 61.2%
  3. 3 Gemma 2 9B 61%
  4. 4 Llama 3.1-8B 47.7%
  5. 5 Mistral 7B v0.1 39.7%

Scores at each effort level

Models the leaderboard ran at two or more effort levels. We don’t have enough cost data here to say what the extra effort costs.

Lowest effort to highest, light to dark
Gemma 2 27B
0-shot 5-shot
58.8%
Llama 3.1-70B
0-shot 5-shot
57.1%
Gemma 2 9B
0-shot 5-shot
56.3%
Llama 3.1-8B
0-shot 5-shot 20-shot
42.7%
Mistral 7B v0.1
0-shot 5-shot
41.4%

Source: The AfriHate paper

ModelEffortHarnessScoreCost per taskOutput tokens
Gemma 2 27B 20-shot 58.8%
Gemma 2 27B 10-shot 57.9%
Gemma 2 27B 5-shot 57.2%
Llama 3.1-70B 20-shot 57.1%
Llama 3.1-70B 10-shot 56.5%
Gemma 2 9B 20-shot 56.3%
Gemma 2 9B 10-shot 55.6%
Llama 3.1-70B 5-shot 55.2%
Gemma 2 9B 5-shot 52.8%
Llama 3.1-8B 20-shot 42.7%
Mistral 7B v0.1 20-shot 41.4%
Llama 3.1-8B 10-shot 40.7%
Mistral 7B v0.1 10-shot 39.9%
Mistral 7B v0.1 5-shot 38.8%
Llama 3.1-8B 5-shot 38.7%
Gemma 2 27B 0-shot 36.6%
Llama 3.1-70B 0-shot 36.5%
Gemma 2 9B 0-shot 21.3%
Llama 3.1-8B 0-shot 19.3%
Mistral 7B v0.1 0-shot 12.5%

Scores at each effort level

Models the leaderboard ran at two or more effort levels. We don’t have enough cost data here to say what the extra effort costs.

Lowest effort to highest, light to dark
Gemma 2 27B
0-shot 5-shot
58.8%
Llama 3.1-70B
0-shot 5-shot
57.1%
Gemma 2 9B
0-shot 5-shot
56.3%
Llama 3.1-8B
0-shot 5-shot 20-shot
42.7%
Mistral 7B v0.1
0-shot 5-shot
41.4%

Source: The AfriHate paper

ModelEffortHarnessScoreCost per taskOutput tokens
Gemma 2 27B 20-shot 58.8%
Gemma 2 27B 10-shot 57.9%
Gemma 2 27B 5-shot 57.2%
Llama 3.1-70B 20-shot 57.1%
Llama 3.1-70B 10-shot 56.5%
Gemma 2 9B 20-shot 56.3%
Gemma 2 9B 10-shot 55.6%
Llama 3.1-70B 5-shot 55.2%
Gemma 2 9B 5-shot 52.8%
Llama 3.1-8B 20-shot 42.7%
Mistral 7B v0.1 20-shot 41.4%
Llama 3.1-8B 10-shot 40.7%
Mistral 7B v0.1 10-shot 39.9%
Mistral 7B v0.1 5-shot 38.8%
Llama 3.1-8B 5-shot 38.7%
Gemma 2 27B 0-shot 36.6%
Llama 3.1-70B 0-shot 36.5%
Gemma 2 9B 0-shot 21.3%
Llama 3.1-8B 0-shot 19.3%
Mistral 7B v0.1 0-shot 12.5%

Number one over time

Lab claims count from the day the card came out, independent runs from the model’s release.

Google DeepMind
Benchmark paper Not tested since Jan 2025
Gemma 2 27B 58.8%

Source: The AfriHate paper

Scores over time

All the scores we have, one dot each, so a model tested at several settings appears more than once.

Plot with other benchmarks
In the benchmark’s paper Best so far
0% 20% 40% 60% 80% 100% 2024 Near the ceiling (85%) Gemma 2 27B Mistral 7B v0.1 0% 20% 40% 60% 80% 100% 2024 Near the ceiling (85%) Gemma 2 27B Mistral 7B v0.1

Source: Official AfriHate leaderboard

No lab model card reports it yet

We read the headline results of 540 launch posts from 57 labs, and none of them list this benchmark.

More in Multilingual

All 545 in Multilingual
  • MMMLU In use MMLU professionally translated into 14 languages. Multiple-choice accuracy. Claude Mythos Preview 92.7%
  • WMT24++ In use English-to-55-language translation of WMT24 test data. Scored with automatic metrics such as MetricX or xCOMET. K-EXAONE 90.5%
  • MMLU-ProX In use MMLU-Pro translated into up to 29 languages with expert review. Multiple-choice accuracy per language. Qwen3.7-Max 87%
  • CLIcK In use Models answer 1,995 multiple-choice questions on Korean language and culture. Questions come from official Korean exams and textbooks in eleven categories. Solar Open 2 90.7%
  • Global MMLU Lite 3.0 In use Models answer 400 human-translated MMLU questions per language in 23 languages. Version 3.0 adds Czech, Hungarian, Oriya, Slovak and Tajik and updates Italian. Accuracy. Claude Opus 5 95.7%
  • MMMB In use Models answer multiple-choice questions about images in six languages across 15 categories. The set has about 12,000 questions. LFM2.5-VL-3B 83%

Data audit

9 of 10 checks pass, 1 warning. Look first at the Newest score check. The newest came 2 years ago, in July 2024, for Llama 3.1-70B. Every benchmark

Catalog record 2 pass
  • Pass Task access All tasks are public.
  • Pass Confidence Checked against a primary source.
Scores 1 warning 3 pass 1 note
  • Pass Independent runs Scores from the benchmark’s paper (15 entries).
  • Pass Models scored 5 models we track, plus 10 other entries on its leaderboard.
  • Warning Newest score The newest came 2 years ago, in July 2024, for Llama 3.1-70B.
  • Note Frontier models None of the top 10 has a score. The strongest tested is Llama 3.1-70B (#177).
  • Pass Headroom The best is 58.8%, under 85%, so it still separates models.
Leaderboard import 3 pass
  • Pass Importer We read the board on arxiv.org: 20 rows for models we track and 10 other entries at the last read.
  • Pass Last read November 20th, 2026.
  • Pass Read worked The last read worked.
Consistency 1 pass 1 note
  • Note Lab claims No model has both a lab number and an independent run.
  • Pass Known issues No published audit found problems.

How we check