Benchmark Almanac

Gemma 3 4B PT

Gemma 3 4B PT places in the top three on EusExams.

25
Benchmarks with a score
0
Ranked first, out of five or more models
1
In the top three
Google DeepMind Mar 2025

Gemma 3 4B PT

Open weights Small

No Epoch Capabilities Index score yet
  • Alyah 63.2% #7 /16
  • FoQA 70.3% #27 /91
  • GermanQuAD 64% #18 /88

How does it do against its rivals?

Gemma 3 4B PT beats gpt-oss-20b on 6 of the 12 benchmarks they share. Qwen3-14B does best against it, ahead on 7 of 13.

  1. gpt-oss-20b 6 ahead, 6 behind Compare
  2. Qwen3-4B-Thinking-2507 5 ahead, 5 behind Compare
  3. DeepSeek-R1-Distill-Qwen-14B 4 ahead, 4 behind Compare
  4. DeepSeek-R1-Distill-Qwen-32B 4 ahead, 4 behind Compare
  5. Aya Expanse 8B 6 ahead, 7 behind Compare
  6. Qwen3-14B 6 ahead, 7 behind Compare

Sources: Official leaderboards, Epoch AI (CC BY 4.0) and labs’ model cards

Where it ranks

Its best score on each benchmark, and where that ranks among every model with a score there. We rank each model once, at its best setting.

Knowledge and language

BenchmarkScoreRankSource
EusExams 46.6% #3 of 24 Official leaderboard
GuiaCat 63.7% #5 of 35 Official leaderboard
EusReading 38.6% #4 of 24 Official leaderboard
EusTrivia 43.7% #4 of 24 Official leaderboard
EuroEval Faroese Grammatical Error Correction 6.10% #14 of 71 Official leaderboard
GermanQuAD 64% #18 of 88 Official leaderboard
MultiBLiMP 85.8% #7 of 25 Official leaderboard
FoQA 70.3% #27 of 91 Official leaderboard
Alyah 63.2% #7 of 16 Official leaderboard
BLiMP-NL 31.1% #10 of 21 Official leaderboard
Dutch CoLA 31.1% #10 of 21 Official leaderboard
FoSent 44% #78 of 147 Official leaderboard
Open Hebrew LLM Leaderboard 48.1 #47 of 88 Official leaderboard
HeQ 65.7% #48 of 88 Official leaderboard
Uncheatable Eval 7.53 #20 of 33 Official leaderboard
GreekMMLU 64.5% #21 of 34 Official leaderboard
QIMMA 45.4% #29 of 44 Official leaderboard
Dutch Book Reviews Dataset 86.1% #70 of 105 Official leaderboard
3LM 62% #17 of 25 Official leaderboard
Global PIQA 58.8% #44 of 59 Official leaderboard
EuroEval ERRNews 22.2 #63 of 76 Official leaderboard
OrangeSum 27.9% #30 of 34 Official leaderboard
NOVA-63 40.6% #46 of 52 Official leaderboard

Safety and security

BenchmarkScoreRankSource
CaBBQ 45% #10 of 11 Official leaderboard
EsBBQ 41.5% #10 of 11 Official leaderboard

Who measured it?

Independent runs cover all 25 of its benchmarks.

  • An independent run only 25
Official leaderboards scored it on 25.

Sources: Official leaderboards, Epoch AI (CC BY 4.0) and labs’ model cards