Gemma 3 4B IT — Benchmarks

Benchmark scores for Gemma 3 4B IT aggregated from public leaderboards (6 boards), with how it ranks among open models and how it compares with its nearest neighbours. See hardware requirements for what you need to run it.

llmrun Composite Score →

64.0

90% range 56.8–71.3 · from 5 boards

CodingNot enough data yet
AgentsNot enough data yet
MathNot enough data yet
ScienceNot enough data yet
Reasoning68.690% range 55–82

Gemma 3 4B IT reasoning benchmarks

Multi-step logic and problem-solving — can the model think through a hard problem rather than recall an answer.

GPQA Diamond
GPQA Diamond is a set of extremely hard, graduate-level science questions (physics, chemistry, biology) written by domain experts and filtered so that skilled non-experts with web access still fail. It measures genuine reasoning rather than memorization.
23.2%
23.2%
median 54.0%open leader 93.1%
Overall best: Gemini 4 Argon (high) 96.4%
5th percentile · #93 of 98 open · #294 of 300 all
#93 of 98
#294 of 300
Independent
Epoch AI · Mar 2025
DTBench
DTBench is a set of handcrafted, expert-validated multiple-choice questions on decision theory — Newcomb-like problems and related puzzles — built by the Conceptual Reasoning Index project with Anthropic. It tests whether a model can follow an abstract argument rather than recall a fact.
50.9%
50.9%
median 67.7%open leader 93.9%
Overall best: Claude Opus 5.5 Max 98.9%
14th percentile · #58 of 68 open · #196 of 214 all
#58 of 68
#196 of 214
Lab-reported
Epoch AI · Mar 2025
LMCA
LMCA (Language Model Conceptual Argumentation) scores models on expert-rated arguments across philosophy, decision theory, and AI risk — domains where there is no empirical feedback loop to check an answer against. It is part of the Conceptual Reasoning Index, built with Anthropic by researchers Emery Cooper and Caspar Oesterheld.
2.8%
2.8%
median 24.7%open leader 55.5%
Overall best: Claude Opus 5.5 Max 68.2%
lowest · #54 of 54 open · #176 of 176 all
#54 of 54
#176 of 176
Lab-reported
Epoch AI · Mar 2025

Gemma 3 4B IT math benchmarks

Symbolic and quantitative problem-solving, from competition math to multi-step calculation.

AIME 2024/2025
AIME (American Invitational Mathematics Examination) is a prestigious high-school competition of hard, integer-answer problems. It's a widely-cited yardstick for multi-step mathematical reasoning, here on the 2024–2025 papers.
7.5%
7.5%
median 52.8%open leader 98.6%
Overall best: Claude Fable 5 (high) 100.0%
26th percentile · #64 of 86 open · #240 of 283 all
#64 of 86
#240 of 283
Independent
Epoch AI · Mar 2025

Gemma 3 4B IT knowledge benchmarks

Breadth and depth of factual knowledge across many subjects, paired with the reasoning to apply it.

MMLU-Pro
MMLU-Pro is a harder, cleaned-up successor to MMLU with ten answer choices and more reasoning-heavy questions across 14 subjects, measuring broad knowledge and reasoning together.
43.6%
43.6%
median 52.1%open leader 88.0%
Overall best: Gemini-3.1-Pro 91.2%
37th percentile · #92 of 145 open · #185 of 257 all
#92 of 145
#185 of 257
Independent
TIGER-Lab

Gemma 3 4B IT instruction following benchmarks

How precisely a model obeys explicit, verifiable instructions — format, length, and content constraints — and, more broadly, how much people prefer its responses head-to-head in blind human voting.

LMArena Text
Arena (arena.ai, formerly LMArena / Chatbot Arena) ranks models by millions of blind, head-to-head human votes on real prompts. The Arena Score shown here is a Bradley-Terry rating computed with Arena's default style control, so higher means people preferred that model's answers more often — it's a relative rating, not a percentage, so compare it only to other models on this board.
1303.3
1303.3
median 1321.5open leader 1487.1
Overall best: Gemini 4 Argon (High) 1523.9
44th percentile · #105 of 187 open · #274 of 405 all
#105 of 187
#274 of 405
Arena
Arena · Oct 2026

Score vs VRAM in the 24 GB tier

Gemma 3 4B IT
Other models
Best Score for its VRAM
x: VRAM (GB) · y: llmrun Score

#33 of 35 that fit 24 GB

Each dot is a model at the best quant that fits a 24 GB GPU. The line joins the models with the best Score for their VRAM; a dot below it is beaten by a model that needs no more memory.

Full 24 GB ranking →

Gemma 3 4B IT vs its neighbours

Same boards for the nearest open models per llmrun Score and the best model that fits 24 GB. The highest score in each row is marked.

Board	Gemma 3 4B ITthis model	Qwen3.8 27Bbest in classQ4_K_M · 18 GB	Qwen2.5 7B Instructnearest by ScoreQ4_K_M · 5.8 GB	Llama 3.1 8B Instructnearest by ScoreQ4_K_M · 7.9 GB	Meta Llama 3 8B Instructnearest by ScoreQ4_K_M · 7.9 GB
llmrun Score	64.0	122.1	65.1	62.5	61.7
GPQA Diamond	23.2%	—	35.5%	27.0%	26.1%
DTBench	50.9%	88.0%	47.7%	50.9%	43.9%
LMCA	2.8%	41.4%	6.4%	5.4%	—
AIME 2024/2025	7.5%	—	2.5%	1.7%	1.9%
MMLU-Pro	43.6%	—	—	44.3%	41.0%
LMArena Text	1303.3	1437.8	—	1211.2	1223.7
Compare head to head		Compare	Compare	Compare	Compare

Scores aggregated from public benchmark sources (Epoch AI: 4 boards, Arena: 1 board, TIGER-Lab: 1 board) — each linked from the benchmark pages. llmrun does not run these benchmarks. Percentiles and ranks are among open models in the llmrun catalogue; the median and leader marks refer to the same field.