Benchmark Almanac

EgyHellaSwag

ALLaM-7B-Instruct 40.2% tops its leaderboard Gemma 3 12B 40.2% leads the models we track
Not in model cards

Models select a plausible ending for a situation described in Egyptian Arabic. The evaluation adapts HellaSwag to the dialect.

Released
May 2025
Kind
Reasoning in many languages
Level
Everyday
Scored
Checked automatically
Answers
Multiple choice
Tasks
All public
Measures
Capability
Language
Arabic
In and out
Text → Text
Interaction
Single turn
Refresh
One fixed set
Run by
Anyone
Based on
HellaSwag
Strongest model tested
Qwen3-8B #154 on the capability index
Last model added
April 2025
Scores refreshed
November 20th, 2026
16
Entries in its paper, 5 models we track

ALLaM-7B-Instruct tops the leaderboard with 40.2% Gemma 3 12B leads with 40.2% among models released by April 2025

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 3 12B, at 40.2%. This view shows only models we track, each at its best.

The newest model with a score here came out in April 2025, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:

  1. 1 ALLaM-7B-Instruct Three-shot 40.2%
  2. 2 Gemma 3 12B Three-shot 40.2%
  3. 3 jais-13b-chat Zero-shot 39%
  4. 4 NileChat (3B) Three-shot 38.3%
  5. 5 AceGPT-13B-chat Three-shot 35.1%
  6. 6 ar-stablelm-2-chat Zero-shot 34.8%
  7. 7 Atlas-Chat-9B Three-shot 34.8%
  8. 8 Qwen2.5-7B-Instruct Zero-shot 34.7%
  9. 9 Gemma 3 4B Three-shot 34.3%
  10. 10 AceGPT-7B-chat Zero-shot 33.3%
  11. 11 Qwen3-8B Zero-shot 31.8%
  12. 12 Qwen2.5-3B-Instruct Zero-shot 31.6%
  13. 13 Qwen3-4B Zero-shot 30.3%
  14. 14 Atlas-Chat-2B Zero-shot 29.7%
  15. 15 Llama-3.2-3B-Instruct Zero-shot 29.2%
  16. 16 Qwen3-1.7B Zero-shot 28.4%

Source: The EgyHellaSwag paper

#EntryScoreSettingDateSource
1 ALLaM-7B-Instruct 40.2% Three-shot Official leaderboard
2 Gemma 3 12B 40.2% Three-shot Mar 2025 Official leaderboard
3 jais-13b-chat 39% Zero-shot Official leaderboard
4 NileChat (3B) 38.3% Three-shot Official leaderboard
5 AceGPT-13B-chat 35.1% Three-shot Official leaderboard
6 ar-stablelm-2-chat 34.8% Zero-shot Official leaderboard
7 Atlas-Chat-9B 34.8% Three-shot Official leaderboard
8 Qwen2.5-7B-Instruct 34.7% Zero-shot Official leaderboard
9 Gemma 3 4B 34.3% Three-shot Mar 2025 Official leaderboard
10 AceGPT-7B-chat 33.3% Zero-shot Official leaderboard
11 Qwen3-8B 31.8% Zero-shot Apr 2025 Official leaderboard
12 Qwen2.5-3B-Instruct 31.6% Zero-shot Official leaderboard
13 Qwen3-4B 30.3% Zero-shot Apr 2025 Official leaderboard
14 Atlas-Chat-2B 29.7% Zero-shot Official leaderboard
15 Llama-3.2-3B-Instruct 29.2% Zero-shot Official leaderboard
16 Qwen3-1.7B 28.4% Zero-shot Apr 2025 Official leaderboard
#ModelScoreSettingDateSource
1 Gemma 3 12B 40.2% Three-shot Mar 2025 Official leaderboard
2 Gemma 3 4B 34.3% Three-shot Mar 2025 Official leaderboard
3 Qwen3-8B 31.8% Zero-shot Apr 2025 Official leaderboard
4 Qwen3-4B 30.3% Zero-shot Apr 2025 Official leaderboard
5 Qwen3-1.7B 28.4% Zero-shot Apr 2025 Official leaderboard

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 3 12B
Zero-shot Three-shot
40.2%
Gemma 3 4B
Zero-shot
34.3%
Qwen3-8B
Three-shot
31.8%
Qwen3-4B
Three-shot
30.3%
Qwen3-1.7B
Three-shot
28.4%

Source: The EgyHellaSwag paper

ModelEffortHarnessScoreCost per taskOutput tokens
Gemma 3 12B Three-shot 40.2%
Gemma 3 12B Zero-shot 38.6%
Gemma 3 4B Three-shot 34.3%
Gemma 3 4B Zero-shot 34.2%
Qwen3-8B Zero-shot 31.8%
Qwen3-8B Three-shot 31.7%
Qwen3-4B Zero-shot 30.3%
Qwen3-4B Three-shot 30.2%
Qwen3-1.7B Zero-shot 28.4%
Qwen3-1.7B Three-shot 28.1%

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 3 12B
Zero-shot Three-shot
40.2%
Gemma 3 4B
Zero-shot
34.3%
Qwen3-8B
Three-shot
31.8%
Qwen3-4B
Three-shot
30.3%
Qwen3-1.7B
Three-shot
28.4%

Source: The EgyHellaSwag paper

ModelEffortHarnessScoreCost per taskOutput tokens
Gemma 3 12B Three-shot 40.2%
Gemma 3 12B Zero-shot 38.6%
Gemma 3 4B Three-shot 34.3%
Gemma 3 4B Zero-shot 34.2%
Qwen3-8B Zero-shot 31.8%
Qwen3-8B Three-shot 31.7%
Qwen3-4B Zero-shot 30.3%
Qwen3-4B Three-shot 30.2%
Qwen3-1.7B Zero-shot 28.4%
Qwen3-1.7B Three-shot 28.1%

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 May 2025
Gemma 3 12B 40.2%

Source: The EgyHellaSwag 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
0% 20% 40% 60% 80% 100% 2025 Near the ceiling (85%) Gemma 3 12B 0% 20% 40% 60% 80% 100% 2025 Near the ceiling (85%) Gemma 3 12B

Source: Official EgyHellaSwag 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 19 months ago, in April 2025, for Qwen3-1.7B. 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 (16 entries).
  • Pass Models scored 5 models we track, plus 11 other entries on its leaderboard.
  • Warning Newest score The newest came 19 months ago, in April 2025, for Qwen3-1.7B.
  • Note Frontier models None of the top 10 has a score. The strongest tested is Qwen3-8B (#154).
  • Pass Headroom The best is 40.2%, under 85%, so it still separates models.
Leaderboard import 3 pass
  • Pass Importer We read the board on arxiv.org: 10 rows for models we track and 11 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